Iterative Collaborative Filtering for Sparse Matrix Estimation
Christian Borgs Jennifer Chayes Christina E. Lee
borgs@microsoft.com jchayes@microsoft.com celee@mit.edu
Microsoft Research New England
One Memorial Drive, Cambridge MA, 02142
Devavrat Shah
devavrat@mit.edu
Massachusetts Institute of Technology
77 Massachusetts Ave, Cambridge, MA 02139
Abstract
The sparse matrix estimation problem consists of estimating the distribution of an n n matrix Y, from a sparsely observed single instance of this matrix where the entries of Y are independent random variables. This captures a wide array of problems; special instances include matrix completion in the context of rec¬ommendation systems, graphon estimation, and community detection in (mixed membership) stochastic block models. Inspired by classical collaborative filtering for recommendation systems, we propose a novel iterative, collaborative filtering-style algorithm for matrix estimation in this generic setting. We show that the mean squared error (MSE) of our estimator goes to 0 as long as ω(d2n) random entries from a total of n2 entries of Y are observed (uniformly sampled), E[Y] has rank d, and the entries of Y have bounded support. The maximum squared error across all entries converges to 0 with high probability as long as we observe a little more, Ω(d2n ln2(n)) entries. Our results are the best known sample complexity results in this generality. Our intuitive, easy to implement iterative nearest-neighbor style algorithm matches the conjectured sample complexity lower bound of d2n for a computationally efficient algorithm for detection in the mixed membership stochastic block model.
1 Introduction
In this work, we propose and analyze an iterative similarity-based collaborative filtering algorithm for the sparse matrix completion problem with noisily observed entries. As a prototype for such a problem, consider a noisy observation of a social network where observed interactions are signals of true underlying connections. We might want to predict the probability that two users would choose to connect if recommended by the platform, e.g. LinkedIn. As a second example, consider a recommendation system where we observe movie ratings provided by users, and we may want to predict the probability distribution over ratings for specific movie-user pairs. The classical collaborative filtering approach is to compute similarities between pairs of users by comparing their commonly rated movies. For a social network, similarities between users would be computed by comparing their sets of friends. We will be particularly interested in the very sparse case where most pairs of users have no common friends, or most pairs of users have no commonly rated movies; thus there is insufficient data to compute the traditional similarity metrics.
31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.
To overcome this limitation, we propose a novel algorithm which computes similarities iteratively, incorporating information within a larger radius neighborhood. Whereas traditional collaborative filtering learns the preferences of a user through the ratings of her/his “friends”, i.e. users who share similar ratings on commonly rated movies, our algorithm learns about a user through the ratings of the friends of her/his friends, i.e. users who may be connected through an indirect path in the data. For a social network, this intuition translates to computing similarities of two users by comparing the boundary of larger radius neighborhoods of their connections in the network. While an actual implementation of our algorithm will benefit from modifications to make it practical, we believe that our approach is very practical; indeed, we plan to implement it in a corporate setting. Like all such nearest-neighbor style algorithms, our algorithm can be accelerated and scaled to large datasets in practice by using a parallel implementation via an approximate nearest neighbor data structure. In this paper, however, our goal is to describe the basic setting and concept of the algorithm, and provide clear mathematical foundation and analysis. The theoretical results indicate that this method achieves consistency (i.e. guaranteed convergence to the correct solution) for very sparse datasets for a reasonably general Latent Variable Model with bounded entries.
The problems discussed above can be mathematically formulated as a matrix estimation problem, where we observe a sparse subset of entries in an m x n random matrix Y, and we wish to complete or de-noise the matrix by estimating the probability distribution of Yij for all (i, j). Suppose that Yij is categorical, taking values in [k] according to some unknown distribution. The task of estimating the distribution of Yij can be reduced to k 1 smaller tasks of estimating the expectation of a binary data matrix, e.g. Yt where Yijt = I(Yij = t) and E[Yijt] = P(Yij= t). If the matrix that we would like to learn is asymmetric, we can transform it to an equivalent symmetric model by defining a new data
matrix Y' = [ 0 Y ~. Therefore, for the remainder of the paper, we will assume a n x n symmetric Y 0
matrix which takes values in [0, 1] (real-valued or binary), but as argued above, our results apply more broadly to categorical-valued asymmetric matrices. We assume that the data is generated from a Latent Variable Model in which latent variables θ1, . . . , θn are sampled independently from U [0, 1], and the distribution of Yij is such that E[Yijθi, θj] = f (θi, θj) =_ Fij for some latent function f. Our goal is to estimate the matrix F. It is worth remarking that the Latent Variable Model is a canonical representation for exchangeable arrays as shown by Aldous and Hoover [5, 25, 7].
We present a novel algorithm for estimating F = [Fij] from a sparsely sampled dataset Yij(i,j)where £ C [n] x [n] is generated by assuming each entry is observed independently with probability p. We require that the latent function f when regarded as an integral operator has finite spectrum with rank d. We prove that the mean squared error (MSE) of our estimates converges to zero at a rate of O((pn)1/5) as long as the sparsity p = ω(d2n1) (i.e. ω(d2n) total observations). In addition, with high probability, the maximum squared error converges to zero at a rate of O((pn)1/5) as long as the sparsity p = Ω(d2n1 ln2(n)). Our analysis applies to a generic noise setting as long as Yij has bounded support. Somewhat surprisingly, our simple nearest-neighbor style algorithm matches the conjectured sample complexity lower bound of total of d2n samples for a computationally efficient algorithm, arising in the context of the mixed membership stochastic block model for detection (weaker than MSE going to 0).
Our work takes inspiration from [1, 2, 3], which estimates clusters of the stochastic block model by computing distances from local neighborhoods around vertices. We improve upon their analysis to provide MSE bounds for the general latent variable model with finite spectrum, which includes a larger class of generative models such as mixed membership stochastic block models, while they consider the stochastic block model with non-overlapping communities. We show that our results hold even when the rank d increases with n, as long as d = o((pn)1/2). As compared to spectral methods such as [28, 39, 20, 19, 18], our analysis handles the general bounded noise model and holds for sparser regimes, only requiring p = ω(n1).
Related work. The matrix estimation problem introduced above includes as specific cases problems from different areas of literature: matrix completion popularized in the context of recommendation systems, graphon estimation arising from the asymptotic theory of graphs, and community detection using the stochastic block model or its generalization known as the mixed membership stochastic block model. The key representative results for each of these are mentioned in Table 1. We discuss the scaling of the sample complexity with respect to d (model complexity, usually rank) and n for polynomial time algorithms, including results for both mean squared error convergence, exact recovery in the noiseless setting, and convergence with high probability in the noisy setting. As can
2
Table 1: Sample Complexity of Related Literature grouped in sections according to the following areas —matrix completion, 1-bit matrix completion, stochastic block model, mixed membership stochastic block model, graphon estimation, and our results
Paper Sample Complexity Data/Noise
[27] ω (dn) noiseless
[28] Ω(dn max(log n, d)), ω (dn) iid Gaussian
[37] ω (dn log n) iid Gaussian
[19] Ω(n max(d, log2 n)) iid Gaussian
[18] ω (dn logs n) indep bounded
[32] Ω(n3/2) iid bounded
[17] Ω(dn log2 n max(d, logo n)) noiseless
[27] Ω(dn max(d, log n)) noiseless
[39] Ω(dn log2 n) noiseless
[19] Ω(n max(d log n, log2 n, d2)) binary entries
[20] Ω(n max(d, log n)), ω (dn) binary entries
[1, 3] ω(d2n) binary entries
[1] Ω(dn log n) binary entries
[6] Ω(d2n polylog n) binary entries
[40] Ω(d2n) binary entries
[4] Ω(n2) binary entries
[43] Ω(n2) binary entries
[10] ω (n) binary entries
this ω(d2n) indep bounded
work Ω(d2n log2 n) indep bounded
Expected matrix Guarantee
rank d MSE-+ 0
rank d MSE-+ 0
rank d MSE-+ 0
rank d MSE-+ 0
rank d MSE-+ 0
Lipschitz MSE-+ 0
rank d exact recovery
rank d exact recovery
rank d exact recovery
rank d MSE-+ 0
rank d MSE-+ 0
d blocks partial recovery
d blocks (SBM) exact recovery
rank d whp error -+ 0
rank d detection
monotone row sum MSE-+ 0
piecewise Lipschitz MSE-+ 0
monotone row sum MSE-+ 0
rank d, Lipschitz MSE-+ 0
rank d, Lipschitz whp error -+ 0
be seen from Table 1, our result provides the best sample complexity for the general matrix estimation problem with bounded entries noise model and rank d, as the other models either require extra log factors, or impose additional requirements on the noise model or the expected matrix. Similarly, ours is the best known sample complexity for high probability max-error convergence to 0 for the general rank d bounded entries setting, as other results either assume block constant or noiseless.
It is worth comparing our results with the known lower bounds on the sample complexity. For the special case of matrix completion with an additive noise model, i.e. Yij = E[Yij] + 77ij and 77ij are i.i.d. zero mean, [16, 20] showed that ω (dn) samples are needed for a consistent estimator, i.e. MSE convergence to 0, and [17] showed that dn log n samples are needed for exact recovery. There is a conjectured computational lower bound for the mixed membership stochastic block model of d2n even for detection, which is weaker than MSE going to 0. Recently, [40] showed a partial result that this computational lower bound holds for algorithms that rely on fitting low-degree polynomials to the observed data. Given that these lower bounds apply to special cases of our setting, it seems that our result is nearly optimal if not optimal in terms of its dependence on both n and d for MSE convergence as well as high probability (near) exact recovery.
Next we provide a brief overview of prior works reported in the Tables 1. In the context of matrix completion, there has been much progress under the low-rank assumption. Most theoretically founded methods are based on spectral decompositions or minimizing a loss function with respect to spectral constraints [27, 28, 15, 17, 39, 37, 20, 19, 18]. A work that is closely related to ours is by [32]. It proves that a similarity based collaborative filtering-style algorithm provides a consistent estimator for matrix completion under the generic model when the latent function is Lipschitz, not just low rank; however, it requires ˜O(n3/2) samples. In a sense, ours can be viewed as an algorithmic generalization of [32] that handles the sparse sampling regime and a generic noise model. Most of the results in matrix completion require additive noise models, which do not extend to setting when the observations are binary or quantized. The USVT estimator is able to handle general bounded noise, although it requires a few log factors more in its sample complexity [18]. Our work removes the extra log factors while still allowing for general bounded noise.
3
There is also a significant amount of literature which looks at the estimation problem when the data matrix is binary, also known as 1-bit matrix completion, stochastic block model (SBM) parameter estimation, or graphon estimation. The latter two terms are found within the context of community detection and network analysis, as the binary data matrix can alternatively be interpreted as the adjacency matrix of a graph – which are symmetric, by definition. Under the SBM, each vertex is associated to one of d community types, and the probability of an edge is a function of the community types of both endpoints. Estimating the n n parameter matrix becomes an instance of matrix estimation. In SBM, the expected matrix is at most rank d due to its block structure. Precise thresholds for cluster detection (better than random) and estimation have been established by [1, 2, 3]. Our work, both algorithmically and technically, draws insight from this sequence of works, extending the analysis to a broader class of generative models through the design of an iterative algorithm, and improving the technical results with precise MSE bounds.
The mixed membership stochastic block model (MMSBM) allows each vertex to be associated to a length d vector, which represents its weighted membership in each of the d communities. The probability of an edge is a function of the weighted community memberships vectors of both endpoints, resulting in an expected matrix with rank at most d. Recent work by [40] provides an algorithm for weak detection for MMSBM with sample complexity d2n, when the community membership vectors are sparse and evenly weighted. They provide partial results to support a conjecture that d2n is a computational lower bound, separated by a gap of d from the information theoretic lower bound of dn. This gap was first shown in the simpler context of the stochastic block model [21]. Our results also achieve this conjectured lower bound, with a sample complexity of ω(d2n) in order to guarantee consistency, which is much stronger than weak detection.
Graphon estimation extends SBM and MMSBM to the generic Latent Variable Model where the probability of an edge can be any measurable function f of real-valued types (or latent variables) associated to each endpoint. Graphons were first defined as the limiting object of a sequence of large dense graphs [14, 22, 34], with recent work extending the theory to sparse graphs [12, 13, 11, 41]. In the graphon estimation problem, we would like to estimate the function f given an instance of a graph generated from the graphon associated to f. [23, 29] provide minimax optimal rates for graphon estimation; however a majority of the proposed estimators are not computable in polynomial time, since they require optimizing over an exponentially large space (e.g. least squares or maximum likelihood) [42, 10, 9, 23, 29]. [10] provided a polynomial time method based on degree sorting in the special case when the expected degree function is monotonic. To our knowledge, existing positive results for sparse graphon estimation require either strong monotonicity assumptions [10], or rank constraints as assumed in the SBM, the 1-bit matrix completion, and in this work.
We call special attention to the similarity based methods which are able to bypass the rank constraints, relying instead on smoothness properties of the latent function f (e.g. Lipschitz) [43, 32]. They hinge upon computing similarities between rows or columns by comparing commonly observed entries. Similarity based methods, also known in the literature as collaborative filtering, have been successfully employed across many large scale industry applications (Netflix, Amazon, Youtube) due to its simplicity and scalability [24, 33, 30, 38]; however the theoretical results have been relatively sparse. These recent results suggest that the practical success of these methods across a variety of applications may be due to its ability to capture local structure. A key limitation of this approach is that it requires a dense dataset with sufficient entries in order to compute similarity metrics, requiring that each pair of rows or columns has a growing number of overlapped observed entries, which does not hold when p = o(n1/2). This work overcomes this limitation in an intuitive and simple way; rather than only considering directly overlapped entries, we consider longer “paths” of data associated to each row, expanding the set of associated datapoints until there is sufficient overlap. Although we may initially be concerned that this would introduce bias and variance due to the sparse sampling, our analysis shows that in fact the estimate does converge to the true solution.
The idea of comparing vertices by looking at larger radius neighborhoods was introduced in [1], and has connections to belief propagation [21, 3] and the non-backtracking operator [31, 26, 36, 35, 8]. The non-backtracking operator was introduced to overcome the issue of sparsity. For sparse graphs, vertices with high-degree dominate the spectrum, such that the informative components of the spectrum get hidden behind the high degree vertices. The non-backtracking operator avoids paths that immediately return to the previously visited vertex in a similar manner as belief propagation, and its spectrum has been shown to be more well-behaved, perhaps adjusting for the high degree vertices, which get visited very often by paths in the graph. In our algorithm, the neighborhood paths
4
are defined by first selecting a rooted tree at each vertex, thus enforcing that each vertex along a path in the tree is unique. This is important in our analysis, as it guarantees that the distribution of vertices at the boundary of each subsequent depth of the neighborhood is unbiased, since the sampled vertices are freshly visited.
2 Model
We shall use graph and matrix notations in an interchangeable manner. For each pair of vertices (i.e. row or column indices) u, v [n], let Yuv [0, 1] denote its random realization. Let denote the edges. If (u, v) , Yuv is observed; otherwise it is unknown.
• Each vertex u [n] is associated to a latent variable θu U[0, 1] sampled i.i.d.
• For each (u, v) [n] [n], Yuv = Yvu [0, 1] is a bounded random variable. Conditioned on θii[n], the random variables Yuv1<u<v<n are independent.
• Fuv := E [Yuv θww[n]] = f(θu, θv) [0, 1] for a symmetric L-Lipschitz function f.
• The function f, when regarded as an integral operator, has finite spectrum with rank d. That is,
f (θu, θv) = Edk=1 λkqk(θu)qk(θv),
where qk are orthonormal L2-integrable basis functions. We assume that there exists some B such that qk (y) B for all k and y [0, 1].
• For every (unordered) index pair (u, v), the entry is observed independently with probability p, i.e. (u, v) and Muv = Mvu = Yuv. If (u, v) / , then Muv = 0.
The data (, M) can be viewed as a weighted undirected graph over n vertices with each (u, v) having weights Muv. The goal is to estimate the matrix F = [Fuv]u,v[n]. Let Λ denote the d d diagonal matrix with λkk[d] as the diagonal entries. Let the eigenvalues be sorted in such a way that λ1λ2•••λd > 0. Let Q denote the d n matrix where Q(k, u) = qk(θu). Since Q is a random matrix depending on the sampled θ, it is not guaranteed to be an orthonormal matrix (even though qk are orthonormal functions). By definition, it follows that F = QT ΛQ. Let d' be the
number of distinct valued eigenvalues. Let Λ˜ denote be the d d' matrix where ˜Λ(a, b) = λa1
b .
Discussing Assumptions. The latent variable model imposes a natural and mild assumption, as Aldous and Hoover proved that if the network is exchangeable, i.e. the distribution over edges is invariant under permutations of vertex labels, then the network can be equivalently represented by a latent variable model [5, 25, 7]. Exchangeability is reasonable for anonymized datasets for which the identity of entities can be easily renamed. Our model additionally requires that the function is L-Lipschitz and has finite spectrum when regarded as an integral operator, i.e. F is low rank; this includes interesting scenarios such as the mixed membership stochastic block model and finite degree polynomials. We can also relax the condition to piecewise Lipschitz, as we only need to ensure that for every vertex u there are sufficiently many vertices v which are similar in function value to u. We assume observations are sampled independently with probability p; however, we discuss a possible solution for dealing with non-uniform sampling in Section 5.
3 Algorithm
The algorithm that we propose uses the concept of local approximation, first determining which datapoints are similar in value, and then computing neighborhood averages for the final estimate. All similarity-based collaborative filtering methods have the following basic format:
1. Compute distances between pairs of vertices, e.g.,
dist(u, a) f0 1 (f (θu, t) f (θa, t))2dt. (1)
2. Form estimate by averaging over “nearby” datapoints,
~
ˆFuv = 1
(a,b) Mab, (2)
where uv := (a, b) s.t. dist(u, a) < ηn, dist(v, b) < ηn.
5
The choice of ηn = (c1pn)1/5 will be small enough to drive the bias to zero, ensuring the included datapoints are close in value, yet large enough to reduce the variance, ensuring Euv diverges.
Inutition. Various similarity-based algorithms differ in the distance computation (Step 1). For dense datasets, i.e. p = ω(n1/2), previous works have proposed and analyzed algorithms which approximate the L2 distance of (1) by using variants of the finite sample approximation,
dist(u, a) = 1
y (Fuy Fay)2, (3)
where y ua iff (u, y) and (a, y) [4, 43, 32]. For sparse datasets, with high probability, ua = for almost all pairs (u, a), such that this distance cannot be computed.
In this paper we are interested in the sparse setting when p is significantly smaller than n1/2, down to the lowest threshold of p = ω(n1). If we visualize the data via a graph with edge set E, then (3) corresponds to comparing common neighbors of vertices u and a. A natural extension when u and a have no common neighbors, is to instead compare the r-hop neighbors of u and a, i.e. vertices y which are at distance exactly r from both u and a. We compare the product of weights along edges in the path from u to y and a to y respectively, which in expectation approximates
[0,1]1 f(θu, t1)(rs=1 2 f(ts, ts+1))f(tr1, θy)d~t = k λr kqk(θu)qk(θy) =eT u QTΛrQey. (4)
We choose a large enough r such that there are sufficiently many “common” vertices y which have paths to both u and a, guaranteeing that our distance can be computed from a sparse dataset.
Algorithm Details. We present and discuss details of each step of the algorithm, which primarily involves computing pairwise distances (or similarities) between vertices.
Step 1: Sample Splitting. We partition the datapoints into disjoint sets, which are used in different steps of the computation to minimize correlation across steps for the analysis. Each edge in is independently placed into 1, 2, or 3, with probabilities c1, c2, and 1 c1 c2 respectively. Matrices M1, M2, and M3 contain information from the subset of the data in M associated to 1, 2, and 3 respectively. M1 is used to define local neighborhoods of each vertex, M2 is used to compute similarities of these neighborhoods, and M3 is used to average over datapoints for the final estimate in (2).
Step 2: Expanding the Neighborhood. We first expand local neighborhoods of radius r around each vertex. Let u,s denote the set of vertices which are at distance s from vertex u in the graph defined by edge set 1. Specifically, i u,s if the shortest path in 1 = ([n], 1) from u to i has a length of s. Let u denote a breadth-first tree in 1 rooted at vertex u. The breadth-first property ensures that the length of the path from u to i within u is equal to the length of the shortest path from u to i in 1. If there is more than one valid breadth-first tree rooted at u, choose one uniformly at random. Let Nu,r [0,1]n denote the following vector with support on the boundary of the r-radius neighborhood of vertex u (we also call Nu,r the neighborhood boundary):
(a,b)pathT (u,i) M1 (a, b) if i u,r,
Nu,r (i) = 0 if i / Su,r,
where path(u, i) denotes the set of edges along the path from u to i in the tree u. The sparsity of Nu,r(i) is equal to u,r, and the value of the coordinate Nu,r(i) is equal to the product of weights along the path from u to i. Let ˜Nu,r denote the normalized neighborhood boundary such that
˜Nu,r = Nu,r/u,r. We will choose radius r to be r = 6 ln(1/p)
8 ln(c1pn) .
Step 3: Computing the distances. For each vertex, we present two variants for estimating the distance.
1. For each pair (u, v), compute dist1(u, v) according to
1c1 p ˜Nu,r ˜Nv,rT M2 ˜Nu,r +1 ˜Nv,r +1. c2p
2. For each pair (u, v), compute distance according to
dist2(u, v) = i[d,] ziΔuv(r, i),
6
where Δuv(r, i) is defined as
~
1c1p~~ ˜Nu,r ˜Nv,r~T M2 ~ ˜Nu,r+i ˜Nv,r+i~, c2p and z Rd' is a vector that satisfies Λ2r +2˜ΛT z = Λ21. z always exists and is unique
because ˜ΛT is a Vandermonde matrix, and Λ2r1 lies within the span of its columns. Computing dist1 does not require knowledge of the spectrum of f. In our analysis we prove that the expected squared error of the estimate computed in (2) using dist1 converges to zero with n for p = ω(n1+) for some > 0, i.e. p must be polynomially larger than n1. Although computing dist2 requires knowledge of the spectrum of f to determine the vector z, the expected squared error of the estimate computed in (2) using dist2 conveges to zero for p = ω(n1), which includes the sparser settings when p is only larger than n1 by polylogarithmic factors. It seems plausible that the technique employed by [2] could be used to design a modified algorithm which does not need to have prior knowledge of the spectrium. They achieve this for the stochastic block model case by bootstrapping the algorithm with a method which estimates the spectrum first and then computes pairwise distances with the estimated eigenvalues.
Step 4: Averaging datapoints to produce final estimate. The estimate Fˆ(u, v) is computed by averaging over nearby points defined by the distance estimates dist1 (or dist2). Recall that B 1 was assumed in the model definition to upper bound supy[0,1] qk(y).
Let uv1 denote the set of undirected edges (a, b) such that (a, b) 3 and both dist1(u, a) and dist1(v, b) are less than ξ1(n) = (c1pn)1/5. The final estimate Fˆ(u, v) produced by using dist1 is computed by averaging over the undirected edge set uv1,
Fˆ(u, v) = 1 E M3(a, b). (5)
uv1 (a,b) 1
Let uv2 denote the set of undirected edges (a, b) such that (a, b) 3, and both dist2(u, a) and dist2(v, b) are less than ξ2(n) = (c1pn)1/5. The final estimate Fˆ(u, v) produced by using dist2 is computed by averaging over the undirected edge set uv 2,
Fˆ(u, v) = 1 E M3(a, b). (6)
uv2 (a,b) 2
4 Main Results
We prove bounds on the estimation error of our algorithm in terms of the mean squared error (MSE),
[1 ~
MSE := E u=v(ˆFuv Fuv)2]
n(n1),
which averages the squared error over all edges. It follows from the model that
.f01 (f (θu, y) f (θv, y))2dy = Edk =1 λ2k(qk(θu) qk(θv))2 = kΛQ(eu ev)k22.
The key part of the analysis is to show that the computed distances are in fact good estimates of ΛQ(eu ev)22. The analysis essentially relies on showing that the neighborhood growth around a vertex behaves according to its expectation, according to some properly defined notion. The radius r must be small enough to guarantee that the growth of the size of the neighborhood boundary is exponential, increasing at a factor of approximately c1pn. However, if the radius is too small, then the boundaries of the respective neighborhoods of the two chosen vertices would have a small intersection, so that estimating the similarities based on the small intersection of datapoints would result in high variance. Therefore, the choice of r is critical to the algorithm and analysis. We are able to prove bounds on the squared error when r is chosen to satisfy the following conditions:
) )
r + d 7 ln(1/c1p) (ln(1/c1p)
8 ln(9c1pn/8) = Θ , r + 1
ln(c1pn) 2 8 ln(7λ2c1pn/8λ1) = Θ
6 ln(1/p) ( ln(1/p) . (7)
ln(c1pn)
The parameter d denotes the number of distinct valued eigenvalues in the spectrum of f, (λ1 ... λd), and determines the number of different radius “measurements” involved in computing dist2(u, v).
7
Computing dist1(u, v) only involves a single measurement, thus the left hand side of (7) can be reduced to r + 1 instead of r + d. When p is above a threshold, we choose c1 to decrease with n to ensure (7) can be satisfied, sparsifying the edge set 1 used for expanding the neighborhood around a vertex . When the sample probability is polynomially larger than n1, i.e. p = n1+ for some
> 0, these constraints imply that r is a constant with respect to n. However, if p = ˜O(n1), we will need r to grow with n according to a rate of 6 ln(1/p)/8 ln(c1pn).
Theorem 4.1. If p = n1+ for some > 0, with a choice of c1 such that c1pn =
( )
Θ max(pn, (p6n7) 119 ), there exists a constant r (with respect to n) which satisfies (7). If
d = o((c1pn)1/2), then the estimate computed using dist1 with parameter r achieves
( (
MSE = O λd2r (c1pn)1/5) = O (c1pn)1/5) .
( ))
With probability greater than 1 O d exp ( (c1pn)12
9B2d , the estimate satisfies
Fˆ Fmax := max
i
,j ˆFij Fij = O(λdr(c1pn)1/10).
Theorem 4.1 proves that the mean squared error (MSE) of the estimate computed with dist1 is bounded by O(λd2r(c1pn)1/5). Therefore, our algorithm with dist1 provides a consistent estimate when r is constant with respect to n, which occurs for p = n1+ for some > 0. In fact, the reason why the error blows up with a factor of λd 2r is because we compute the distance by summing product of weights over paths of length 2r. From (4), we see that in expectation, when we take the product of edge weights over a path of length r from u to y, instead of computing f(θu, θy) = eTu QΛQey, the expression concentrates around eTu QΛrQey, which contains extra factors of Λr1. Therefore, by computing over a radius r, the calculation in dist1 will approximate kΛr +1Q(eu ev)k22 rather than our intended kΛQ(eu ev)k22, thus leading to an error factor of λd2r. It turns out that dist2 adjusts for this bias, as the multiple measurements Δuv(r, i) with different length paths allows us to separate out ekΛQ(eu ev) for all k with distinct values of λk.
( )
Theorem 4.2. Ifp = O(n2/3), with a choice of c1 such that c1pn = Θ max(pn, (p6n7) 1
(8'+11) ) ,
there exists a value for r which satisfies (7). If d = o((c1pn)1/2) and d = o(r), then the estimate computed using dist2 with parameter r achieves
(MSE = O (c1pn)1/5).
( ))
If p = Ω(n1d2 ln2(n)), with probability 1 O d exp ( (c1pn)12
9B2d , the estimate satisfies
Fˆ Fmax := max
i
,j ˆFij Fij = O((c1pn)1/10).
Theorem 4.2 proves that the mean squared error (MSE) of the estimate computed using dist2 is
bounded by O((c1pn)1/5); and thus the estimate is consistent in the ultra sparse sampling regime of p = ω(d2n1). We also present high probability bounds on the squared error of each entry.
Lemma 4.3. For any u, v [n], if d = o((c1pn)1/2), with probability at least
(
1 O d exp ( (c1pn)12 ~ + exp ~ c3pn2(c1pn)25 ~)
8B2d 48L2λ12,
the squared error of the estimate computed with dist1 for parameter r satisfying (7) is bounded by ( ˆFuv f (θu, θv))2 = O(λd2r(c1pn)1/5).
Lemma 4.4. For any u, v [n], assuming d = o((c1pn)1/2) and d = o(r), with probability at least
( ( ))
1 O (c1pn)12
d exp 8B2d
the squared error of the estimate computed with dist2 for parameter r satisfying (7) is bounded by ( ˆFuv f (θu, θv))2 = O((c1pn)1/5).
8
5 Discussion
In this work we presented a similarity based collaborative filtering algorithm which is provably consistent in sparse sampling regimes, as long as the sample probability p = ω(n1). The algorithm computes similarity between two users by comparing their local neighborhoods. Our model assumes that the data matrix is generated according to a latent variable model, in which the weight on an observed edge (u, v) is equal in expectation to a function f over associated latent variables 0u and 0v. We presented two variants for computing similarities (or distances) between vertices. Computing dist1 does not require knowledge of the spectrum of f, but the estimate requires p to be polynomially larger than n in order to guarantee the expected squared error converges to zero. Computing dist2 uses the knowledge of the spectrum of f, but it provides an estimate that is provably consistent for a significantly sparse regime, only requiring that p = ω(n1). The mean squared error of both algorithms is bounded by O((pn)1/5). Since the computation is based on of comparing local neighborhoods within the graph, the algorithm can be easily implemented for large scale datasets where the data may be stored in a distributed fashion optimized for local graph computations.
Practical implementation. In practice, we do not know the model parameters, and we would use cross validation to tune the radius r and threshold ηn. If r is either too small or too large, then the vector Nu,r will be too sparse. The threshold ηn trades off between bias and variance of the final estimate. Since we do not know the spectrum, dist1 may be easier to compute, and still enjoys good properties as long as r is not too large. When the sampled observations are not uniform across entries, the algorithm may require more modifications to properly normalize for high degree hub vertices, as the optimal choice of r may differ depending on the local sparsity. The key computational step of our algorithm involves comparing the expanded local neighborhoods of pairs of vertices to find the “nearest neighbors”. The local neighborhoods can be computed in parallel, as they are independent computations. Furthermore, the local neighborhood computations are suitable for systems in which the data is distributed across different machines in a way that optimizes local neighborhood queries. The most expensive part of our algorithm involves computing similarities for all pairs of vertices in order to determine the set of nearest neighbors. However, it would be possible to use approximate nearest neighbor techniques to greatly reduce the computation such that approximate nearest neighbor sets could be computed with significantly fewer than n2 pairwise comparisons.
Non-uniform sampling. In reality, the probability that entries are observed is not be uniform across all pairs (i, j). However, we believe that an extension of our result can also handle variations in the sample probability as long as the sample probability is a function of the latent variables and scales in the same way with respect to n across all entries. Suppose that the probability of observing (i, j) is given by pg(0i, 0j), where p is the scaling factor (contains the dependence upon n), and g allows for constant factor variations in the sample probability across entries as a function of the latent variables. If we let matrix X indicate the presence of an observation or not, then we can apply our algorithm twice, first on matrix X to estimate function g, and then on data matrix M to estimate f times g. We can simply divide by the estimate for g to obtain the estimate for f. The limitation is that if g(0i, 0j) is very small, then the error in estimating the corresponding f (0i, 0j) will have higher variance. However, it is expected that error increases for edge types with fewer samples.
Acknowledgments
This work is supported in parts by NSF under grants CMMI-1462158 and CMMI-1634259, by DARPA under grant W911NF-16-1-0551, and additionally by a NSF Graduate Fellowship and Claude E. Shannon Research Assistantship.
References
[1] Emmanuel Abbe and Colin Sandon. Community detection in general stochastic block models: Fundamental limits and efficient algorithms for recovery. In Foundations of Computer Science (FOCS), 2015 IEEE 56th Annual Symposium on, pages 670–688. IEEE, 2015.
[2] Emmanuel Abbe and Colin Sandon. Recovering communities in the general stochastic block model without knowing the parameters. In Advances in neural information processing systems, 2015.
9
[3] Emmanuel Abbe and Colin Sandon. Detection in the stochastic block model with multiple clusters: proof of the achievability conjectures, acyclic bp, and the information-computation gap. Advances in neural information processing systems, 2016.
[4] Edo M Airoldi, Thiago B Costa, and Stanley H Chan. Stochastic blockmodel approximation of a graphon: Theory and consistent estimation. In Advances in Neural Information Processing Systems, pages 692–700, 2013.
[5] D.J. Aldous. Representations for partially exchangeable arrays of random variables. J. Multi¬variate Anal., 11:581 – 598, 1981.
[6] Animashree Anandkumar, Rong Ge, Daniel Hsu, and Sham Kakade. A tensor spectral approach to learning mixed membership community models. In Conference on Learning Theory, pages 867–881, 2013.
[7] Tim Austin. Exchangeable random arrays. Technical Report, Notes for IAS workshop., 2012.
[8] Charles Bordenave, Marc Lelarge, and Laurent Massoulié. Non-backtracking spectrum of random graphs: community detection and non-regular ramanujan graphs. In Foundations of Computer Science (FOCS), 2015 IEEE 56th Annual Symposium on, pages 1347–1357. IEEE, 2015.
[9] Christian Borgs, Jennifer Chayes, and Adam Smith. Private graphon estimation for sparse graphs. In Advances in Neural Information Processing Systems, pages 1369–1377, 2015.
[10] Christian Borgs, Jennifer T Chayes, Henry Cohn, and Shirshendu Ganguly. Consistent nonpara-metric estimation for heavy-tailed sparse graphs. arXiv preprint arXiv:1508.06675, 2015.
[11] Christian Borgs, Jennifer T Chayes, Henry Cohn, and Nina Holden. Sparse exchangeable graphs and their limits via graphon processes. arXiv preprint arXiv:1601.07134, 2016.
[12] Christian Borgs, Jennifer T Chayes, Henry Cohn, and Yufei Zhao. An LP theory of sparse graph convergence I: limits, sparse random graph models, and power law distributions. arXiv preprint arXiv:1401.2906, 2014.
[13] Christian Borgs, Jennifer T Chayes, Henry Cohn, and Yufei Zhao. An LP theory of sparse graph convergence II: Ld convergence, quotients, and right convergence. arXiv preprint arXiv:1408.0744, 2014.
[14] Christian Borgs, Jennifer T Chayes, László Lovász, Vera T Sós, and Katalin Vesztergombi. Convergent sequences of dense graphs I: Subgraph frequencies, metric properties and testing. Advances in Mathematics, 219(6):1801–1851, 2008.
[15] Emmanuel Candes and Benjamin Recht. Exact matrix completion via convex optimization. Communications of the ACM, 55(6):111–119, 2009.
[16] Emmanuel J Candes and Yaniv Plan. Matrix completion with noise. Proceedings of the IEEE, 98(6):925–936, 2010.
[17] Emmanuel J Candès and Terence Tao. The power of convex relaxation: Near-optimal matrix completion. IEEE Transactions on Information Theory, 56(5):2053–2080, 2010.
[18] Sourav Chatterjee. Matrix estimation by universal singular value thresholding. The Annals of Statistics, 43(1):177–214, 2015.
[19] Yudong Chen and Martin J Wainwright. Fast low-rank estimation by projected gradient descent: General statistical and algorithmic guarantees. arXiv preprint arXiv:1509.03025, 2015.
[20] Mark A Davenport, Yaniv Plan, Ewout van den Berg, and Mary Wootters. 1-bit matrix completion. Information and Inference, 3(3):189–223, 2014.
[21] Aurelien Decelle, Florent Krzakala, Cristopher Moore, and Lenka Zdeborová. Asymptotic analysis of the stochastic block model for modular networks and its algorithmic applications. Phys. Rev. E, 84:066106, Dec 2011.
10
[22] Persi Diaconis and Svante Janson. Graph limits and exchangeable random graphs. Rendiconti di Matematica, VII(28):33–61, 2008.
[23] Chao Gao, Yu Lu, and Harrison H Zhou. Rate-optimal graphon estimation. The Annals of Statistics, 43(6):2624–2652, 2015.
[24] David Goldberg, David Nichols, Brian M. Oki, and Douglas Terry. Using collaborative filtering to weave an information tapestry. Commun. ACM, 1992.
[25] D.N. Hoover. Row-column exchangeability and a generalized model for probability. In Exchangeability in Probability and Statistics (Rome, 1981), pages 281 – 291, 1981.
[26] Brian Karrer, M. E. J. Newman, and Lenka Zdeborová. Percolation on sparse networks. Phys. Rev. Lett., 113:208702, Nov 2014.
[27] Raghunandan H Keshavan, Andrea Montanari, and Sewoong Oh. Matrix completion from a few entries. IEEE Transactions on Information Theory, 56(6):2980–2998, 2010.
[28] Raghunandan H Keshavan, Andrea Montanari, and Sewoong Oh. Matrix completion from noisy entries. Journal of Machine Learning Research, 11(Jul):2057–2078, 2010.
[29] Olga Klopp, Alexandre B Tsybakov, and Nicolas Verzelen. Oracle inequalities for network models and sparse graphon estimation. To appear in Annals of Statistics, 2015.
[30] Yehuda Koren and Robert Bell. Advances in collaborative filtering. In Recommender Systems Handbook, pages 145–186. Springer US, 2011.
[31] Florent Krzakala, Cristopher Moore, Elchanan Mossel, Joe Neeman, Allan Sly, Lenka Zde-borová, and Pan Zhang. Spectral redemption in clustering sparse networks. Proceedings of the National Academy of Sciences, 110(52):20935–20940, 2013.
[32] Christina E. Lee, Yihua Li, Devavrat Shah, and Dogyoon Song. Blind regression: Nonpara-metric regression for latent variable models via collaborative filtering. In Advances in Neural Information Processing Systems 29, pages 2155–2163, 2016.
[33] Greg Linden, Brent Smith, and Jeremy York. Amazon.com recommendations: Item-to-item collaborative filtering. IEEE Internet Computing, 7(1):76–80, 2003.
[34] László Lovász. Large networks and graph limits, volume 60. American Mathematical Society Providence, 2012.
[35] Laurent Massoulié. Community detection thresholds and the weak ramanujan property. In Proceedings of the Forty-sixth Annual ACM Symposium on Theory of Computing, STOC ’14, pages 694–703, New York, NY, USA, 2014. ACM.
[36] Elchanan Mossel, Joe Neeman, and Allan Sly. A proof of the block model threshold conjecture. Combinatorica, Aug 2017.
[37] Sahand Negahban and Martin J Wainwright. Estimation of (near) low-rank matrices with noise and high-dimensional scaling. The Annals of Statistics, pages 1069–1097, 2011.
[38] Xia Ning, Christian Desrosiers, and George Karypis. Recommender Systems Handbook, chapter A Comprehensive Survey of Neighborhood-Based Recommendation Methods, pages 37–76. Springer US, 2015.
[39] Benjamin Recht. A simpler approach to matrix completion. Journal of Machine Learning Research, 12(Dec):3413–3430, 2011.
[40] David Steurer and Sam Hopkins. Bayesian estimation from few samples: community detection and related problems. https://arxiv.org/abs/1710.00264, 2017.
[41] Victor Veitch and Daniel M Roy. The class of random graphs arising from exchangeable random measures. arXiv preprint arXiv:1512.03099, 2015.
11
[42] Patrick J Wolfe and Sofia C Olhede. Nonparametric graphon estimation. arXiv preprint arXiv:1309.5936, 2013.
[43] Yuan Zhang, Elizaveta Levina, and Ji Zhu. Estimating network edge probabilities by neighbor¬hood smoothing. arXiv preprint arXiv:1509.08588, 2015.
12
Association for Information Systems
AIS Electronic Library (AISeL)
Mediterranean Conference on Information Systems
MCIS 2014 Proceedings
(MCIS)
Summer 9-4-2014
A SWOT ANALYSIS OF THE SOFT
COMPUTING PARADIGMS SOA/SOC/
CLOUD COMBINATION (C-SOA) IN
SOFTWARE DEVELOPMENT
Natalia, Kryvinska
University of Vienna, Vienna, natalia.kryvinska@univie.ac.at
Christine Bauer
Institute for Management Information Systems, Vienna University of Economics and Business, Vienna, Austria.,
chris.bauer@wu.ac.at
Christine Strauss,
University of Vienna, Vienna, christine.strauss@univie.ac.at
Michal, Gregus,
University in Bratislava, Bratislava, Michal.Gregus@fm.uniba.sk
Follow this and additional works at: http://aisel.aisnet.org/mcis2014
Recommended Citation
Kryvinska, Natalia,; Bauer, Christine; Strauss,, Christine; and Gregus,, Michal,, "A SWOT ANALYSIS OF THE SOFT-COMPUTING PARADIGMS SOA/SOC/CLOUD COMBINATION (C-SOA) IN SOFTWARE DEVELOPMENT" in Mola, L., Carugati, A,. Kokkinaki, A., Pouloudi, N., (eds) (2014) Proceedings of the 8th Mediterranean Conference on Information Systems, Verona, Italy, September 03-05. CD-ROM. ISBN: 978-88-6787-273-2.
http://aisel.aisnet.org/mcis2014/27
This material is brought to you by the Mediterranean Conference on Information Systems (MCIS) at AIS Electronic Library (AISeL). It has been accepted for inclusion in MCIS 2014 Proceedings by an authorized administrator of AIS Electronic Library (AISeL). For more information, please contact elibrary@aisnet.org.
A SWOT ANALYSIS OF THE SOFT-COMPUTING
PARADIGMS SOA/SOC/CLOUD COMBINATION (C-SOA) IN
SOFTWARE DEVELOPMENT
Complete Research
Kryvinska, Natalia, University of Vienna, Vienna, Austria, natalia.kryvinska@univie.ac.at
Bauer, Christine, Vienna University of Economics and Business, Vienna, Austria, chris.bauer@wu.ac.at
Strauss, Christine, University of Vienna, Vienna, Austria, christine.strauss@univie.ac.at
Gregus, Michal, Comenius University in Bratislava, Bratislava, Slovakia, Michal.Gregus@fm.uniba.sk
Abstract
Due to their technological complexity, traditional software development paradigms are not appropriate to face the challenges in the modern Web 2.0 world. Having the ability to adapt rapidly to the fast changing Web in an open environment, and challenged by the need for applications to be flexible, Service-Oriented Architecture (SOA), Service-Oriented Computing (SOC), and more recently Cloud Computing are becoming more and more popular. SOC/SOA and Cloud Computing share many drivers, such as enterprise portfolio and cost reduction. Both approaches are complementary and are expected to become the core of IT-based projects and/or businesses. Thus, this paper discusses the opportunities and challenges of Soft-Computing Paradigms, which are evaluated based on a SWOT analysis. For illustration, we also present a case of SOC/Cloud (C-SOA) based on the real-world application of Amazon Services.
Keywords: Cloud Computing, Service-Oriented Architecture (SOA), Service-Oriented Computing (SOC), Strengths Weaknesses Opportunities Threats (SWOT), Web Services, C-SOA.
Eighth Mediterranean Conference on Information Systems, Verona 2014 1
Kryvinska et al. / SWOT Analysis for Soft-Computing
1 Introduction
The shift from traditional to rather novel soft-computing paradigms in recent years has triggered extended discussions among scholars as well as industrial experts (Kryvinska et al., 2011b). Gellersen and Gaedke (1999) claimed that traditional software development approaches typically have four phases: analysis, design, implementation and maintenance/evolution. Since the introduction of the Internet, experts were looking for some kind of structured approach for software development and maintenance for and on the Internet. “The Web is defined as an information medium rather than an application platform” (Huang and Mak, 2003, p. 41) and therefore it is frequently subject to changes. The paradigm of Object-Oriented Computing (OOP) came up in the 1980’s, as a result of this problem. The major advantage of OOP is the sharing and reusing of resources. Rine and Bhargava (1992) claimed that it was the best paradigm available, as software could be maintained longer. OOP enabled to alter and align Web applications faster and better on required changes than older paradigms such as structured programming. However, OOP is rather considered evolutionary than revolutionary, similar to other paradigms before and after OOP (Rine and Bhargava, 1992). This is the main reason why it did not emerge any overall state-of-the-art approach for software development on the Internet; and the development relies heavily on the individual skills of the developers and their best practices (Huang and Mak 2003).
Due to the fact that software engineering has “inherently no silver bullet” (Brooks, 1987, p. 2) in the evolution of technology and business trends, the Service-Oriented Architecture (SOA) and Service-Oriented Computing (SOC) evolved around 2005 (Huhns and Singh, 2005). SOC is a design paradigm to build a composition (in form of computer software) out of independent distributed services in SOA, which consists of three parts: a provider of applications, a consumer, and a registry. In SOA, (Web) services represent business functionalities, principles, concepts, and applications (Auer et al., 2011; Kryvinska et al., 2008). These are built as individual software components, which can be reused for different purposes among different applications. As a consequence, talking about SOC requires also talking about SOA (Huhns and Singh, 2005). Cloud computing, which has become popular in the last few years, shares many of the drivers as in SOC/SOA, such as enterprise portfolio and cost reduction (Mladenow et al., 2012). Raines (2009) states that SOC/SOA and cloud computing are complementary approaches, which bear benefits but also risks. The application fields are manifold and include, for instance, contextual digital signage (Bauer et al., 2011), damage prevention (Strauss et al., 2009), or tracking of high-value items such as arts (Bauer et al., 2013).
Thus, this paper is structured as follows: First we discuss related work on the soft-computing paradigms SOC/SOA and cloud computing. Then we evaluate these approaches based on a SWOT analysis. For an illustration, we then discuss these approaches with an exploration of a real-world example.
2 The Definitions
As it is defined in Huhns and Singh (2005), the architecture of service-based applications contains three main parts: provider, registry/broker, and requestor. These parts interact using publish, find, and bind operations. The service provider is an organization that provides access to a Web service and publishes the service description in a service registry run by a broker. The requestor finds the service description in a service registry and uses the information in the description to bind to a service. In a Service-Oriented Architecture (SOA), the service registry provides a centralized location for storing
Eighth Mediterranean Conference on Information Systems, Verona 2014 2
Kryvinska et al. / SWOT Analysis for Soft-Computing
service descriptions with following key standards: Find - UDDI (Universal Description Discovery and Integration) (Erenkrantz, 2004); Bind - SOAP (Simple Object Access Protocol), simple XML-based protocol to let applications exchange information over HTTP to access a Web services (Erenkrantz, 2004); Publish - WSDL (Web Services Description Language), document written in a XML, specifies the location of the service and the operations the service offers (Wei and Blake, 2010).
“Service-Oriented Computing (SOC) is an emerging computing paradigm whose main goal is to support the development of distributed applications in heterogeneous environments” (Crasso et al., 2011). Deploying SOC enables building software systems by assembling together distributed functionalities.
This new paradigm is rooted in object-oriented and component-based software development. Its primary goal is to enable developers to build networks of interoperable and collaborative applications, regardless of these applications’ programming languages and the platforms they run on. This goal is achieved through the use of independent computational units, which are called “services” (Fantechi et al., 2012).
In addition, SOC brings in software qualities, which are of major importance. With its planned reuse approach, SOC furthers rapid, high quality development of software applications. By using existing, already tested software elements, the time needed to build an application is reduced and its overall quality is improved (Chollet et al., 2012; Arbab, 2012).
SOC uses SOA services to develop applications in the fast changing world of the Internet. These services are independent entities and can be “used in a platform independent way” (Papazoglou et al., 2006). The requester can compound them, as he or she wants, as long as these services follow certain standards. These standards (e.g., UDDI) have been developed by major computing companies (such as BEA, IBM, Microsoft, Amazon, etc.), which have been moved towards the paradigm of SOC and SOA (Tsai et al., 2006).
Cloud computing is booming in recent years, and the IT industry shifts more and more to the cloud, although cloud computing is not a new technology. Rather it gained acceptance in recent years; but actually the cloud-computing paradigm was first mentioned in 1997 (Chellapa, 1997).
Cloud computing is the distribution of applications and/or hardware provided as services over the Internet and in data centers (e.g., servers) (Armbrust et al., 2010; Erdogmus, 2009), and is often referred to as “the Cloud”. Cloud Computing exists in three forms (levels of services): Software as a Service (SaaS), Infrastructure as a Service (IaaS) and Platform as a Service (PaaS) (NIST-2011). Definitions in literature are heterogeneous and the separation lines between the levels of services are drawn differently. Dillon et al (2010) define these three services as follows:
• SaaS (Software as a Service) - Software that is offered over the Internet, available to the end consumer as and when wanted;
• IaaS (Infrastructure as a Service) - Organization outsources the equipment for operations, including storage, hardware, servers and networking;
• PaaS (Platform as a Service) - Organizations provide infrastructure on which software developers can build new applications or extend existing ones.
A federal cloud is defined “as the deployment and management of multiple external and internal cloud computing services to match business needs. A federation is the union of several smaller parts that
Eighth Mediterranean Conference on Information Systems, Verona 2014 3
Kryvinska et al. / SWOT Analysis for Soft-Computing
perform a common action” (Rouse, 2011). As can be seen from Figure 1, hybrid architecture can be used additionally to the federal cloud. Thereby, a service provider may be in a cloud or not, and a SOA/SOC may involve any combination of clouds and non-clouds (Barry and Dick, 2013).
Figure 1. SOA, SOC and Cloud working together (source Barry and Dick, 2013).
Cloud Computing and SOC/SOA are independent approaches; still, they may be used as complementary activities (Figure 2). Cloud Computing is a broad term for any Web service, which offers the entire “traditional IT stack” (Raines, 2009), such as software, hardware, and applications. SOC/SOA, instead, focuses mainly on software services (Raines, 2009; Kryvinska et al., 2011a). In order to be combinable and benefit from each other, both approaches need to follow the Web service standards (i.e., UDDI, WSDL, SOAP).
Eighth Mediterranean Conference on Information Systems, Verona 2014 4
Kryvinska et al. / SWOT Analysis for Soft-Computing
Figure 2. Venn diagram of relationships between Web Services, SOA, and Cloud Computing (source Barry and Dick, 2013).
3 The SWOT Analysis
The idea behind a SOC and Cloud Computing combination (which is referred to as C-SOA) is to make business easier and offer outsourcing solutions for enterprises (Figure 3). However, using SOC within “the cloud” environment does not only bring benefits, but also challenges arise. Based on a SWOT analysis, we evaluate this combination. The SWOT analysis is a strategic planning method (cf. Hill and Westbrook, 1997; Weihrich, 1982). Thereby, SWOT is an acronym representing strength, weaknesses, opportunities, and threats, which characterize the dimensions along with the entities or situations to be analyzed. Based on the analysis for each dimension, respective strategies are derived. Compared to other strategic planning methods, a particular strength of the SWOT analysis is that it considers both internal (strengths, weaknesses) as well as external (opportunities, threats) dimensions (Weihrich, 1982).
Figure 3. Guarding against Cloud silos with C-SOA (source Shaheen, 2012).
Eighth Mediterranean Conference on Information Systems, Verona 2014 5
Kryvinska et al. / SWOT Analysis for Soft-Computing
While strategic planning has a long history in military, predominantly product-oriented companies have extensively used the SWOT analysis (Weihrich, 1982). Only later this analysis’ value was recognized for service-oriented businesses and ideas (e.g., Bernroider, 2002; Andrews, 2009).
The SWOT analysis is summarized in Table 4; details can be found in the following subsections.
Strengths Weaknesses
• integration over time • lack of standardization
• SOA standards available (i.e., UDDI, SOAP, WSDL) •
• lack of interoperability
lack of portability
• rapid services composing and orchestration • communication of software components of
• services re-use several cloud service providers insufficient
• managing services heterogeneity • additional workload
• rapid elasticity • more complicated workflow
• high scalability • “dynamically accessible” versus
• fault tolerance “dynamically discoverable”
• distributed functionality • high availability required
• on-demand self-service • no cloud is 100% reliable
• resources pooling • changing providers may imply
• managing complex software intensive redevelopment of solutions
systems • high dependence on provider
• rapid resources reconfiguration • the service level has not yet evolved
Opportunities Threats
• potential of standards, agreements, and rules • availability of cloud not perfectly 24/7
• federal clouds • constant merging and relocating of
• intermediary layer companies hampers seamless interoperation
• rapid service deployment may increase of web services
interoperability • trust issues
• introduction of a Q-Cloud Service • messages could be intercepted or inferred by
• integration of standards of ontologies in Web services the competition or anybody who can use this information
• dynamic/rapid building complex system-of- systems on demand • security isolation
Table 4. The SWOT Analysis.
SOC/SOA and Cloud computing may be used in combination as complements or in separation as independent solutions. This is a strength as they have not to be introduced at the same time but may be integrated over time (Wei and Blake, 2010).
There are already Open API platforms (e.g., the SUN Open Cloud Platform) that support federal clouds, which define key resources (via HTTP and JSON) like Cloud, Virtual Data Center, Cluster, Virtual Machine, etc., especially when they cooperate with other projects like Eucalyptus (Dillon et al., 2010).
There are already some standards available from SOA (i.e., UDDI, SOAP, WSDL) that C-SOA may build on. The Eucalyptus Project, for example, follows the approach of UDDI in the SOC/SOA paradigm, by starting the partnership with Amazon (Dillon et al., 2010).
Due to the re-use approach, rapid service composing and resources reconfiguration is possible. C-SOA allows for rapid elasticity and high scalability as required by a service consumer, and profits from
Eighth Mediterranean Conference on Information Systems, Verona 2014 6
Kryvinska et al. / SWOT Analysis for Soft-Computing
resources pooling, due to the cloud-computing ingredient. Further strengths are the integration and management of a heterogeneous and complex service landscape.
Overall, we can summarize that C-SOA combines the strengths of SOA and cloud computing, which are widely discussed in scientific literature and do not need further elaboration at this place.
The management of cloud services, which is inherent in C-SOA, implies an additional workload (Tsai et al., 2010) and also concerning the infrastructure, there are certain limitations, which may interfere or complicate the workflow (Tsai et al., 2010).
Currently, Web services are rather “dynamical accessible” than “dynamical discoverable” with the current standards (e.g., WSDL or SOAP). Problems are identified in the detailed information of semantics, which are provided by individuals. For example if someone looks for tax preparation software in the US, and types in a code like “xx.xx.xx.xx.xx”, the right information for offered tax service software could be parsed out or not, depending on the provided semantics (Dogac et al., 2002). Accordingly, while for SOA some standards already exist (i.e., UDDI, SOAP, WSDL), there is a lack of standardization for C-SOA (Wei and Blake, 2010) (Dillon et al., 2010). This weakness also goes in hand with the lack of well-defined semantics in XML and the movement of web services beyond their key standards (UDDI, WSDL and SOAP) (Yu et al., 2008).
As a result of the lack of standardization, the interoperability of the cloud services maintains a problem. Dillon et al. (2010), for instance, argues that SaaS providers use different application domains like ERP, CRM, etc. and experts at KDD09 panel in 2009 (Zeller et al., 2009) raised the issue that data mining, which is also a SaaS, requires standardization to operate properly.
The lack of well-running interoperations and service deployment also cause a lack of portability, as SOC depends on these (Wei and Blake, 2010).
Due to the lack of standardization between service providers, also the communication of software components of several cloud service providers is currently insufficient (Wei and Blake, 2010). Accordingly, there is a need of cloud service providers to unify (Wei and Blake, 2010).
Furthermore, SOA systems frequently require a high availability. However, loud computing is still struggling with server outages, which make the combination of the two challenging. The recent example of Amazon’s EC2 hosting service has shown that not only the cloud provider itself suffers under availability problems, but also companies, which rely on the respective cloud services (Hesseldahl, 2012). In addition, from time to time, server updates of the cloud provider require a shutdown. In short, no cloud is 100% reliable as they suffer from human failures and technical problems (Dillon et al., 2010).
Additionally, cloud solutions are not always designed for cross-cloud usage (Wei and Blake, 2010). The problem is and if an organization changes, it is not unlikely that – the despite a lot of time was spent on developing applications in the cloud –the solution has to be redeveloped at the new cloud provider (Tsai et al., 2010).
As a consequence, to lower the additional costs when moving to another, provider cloud user is often stuck with one provider because the migration from one cloud to another is hard to undertake (Tsai et al., 2010). Accordingly there is strong dependence on a single cloud provider, as only a switch to other solutions of the very same cloud provider seems feasible (Tsai et al., 2010). And even the connectivity between cloud services of the same provider still have to be optimized, which requires novel approaches as current approaches do not suffice (Wei and Blake, 2010).
Eighth Mediterranean Conference on Information Systems, Verona 2014 7
Kryvinska et al. / SWOT Analysis for Soft-Computing
Furthermore, although there is a possibility for organizations to outsource to different cloud service providers, this would, however, cause additional complexity (Dillon et al., 2010) and, thus, would limit the benefits of the C-SOA approach.
Another obstacle is the service level agreement (SLA), as the service level has not yet evolved as far as it needs to be (for customers). Especially if a business is subject to constant changes, this causes problems, as SLA supports tend to be static (Tsai et al., 2010).
The various approaches and models are suggested by scientific literature point to more or less the same opportunity: The potential of standards, agreements, and rules.
Service discovery could, for instance, be realized through federal clouds (Wei and Blake, 2010). In addition, scientific literature suggests also several opportunities with respect to federal clouds:
Integrating an intermediary layer between cloud consumers to manage the various interactions of APIs with providers is one opportunity (Dillon et al., 2010). A prominent example is OpenNebula, which is an open-source project. Its mission is to provide a solution for managing and organizing data centers and IaaS clouds (OpenNebula.org, 2012). Another solution for a layer is to integrate high-level application requirements for low-level cloud resources (Sun et al., 2012). Consumer needs are used to specify high-level application requirements. Then they get translated into high-level infrastructures, which can be translated onto low-level descriptions of cloud resources.
Another opportunity lies in the standardization between cloud providers. To gain positive impact out of federal clouds, companies may unify and standardize (Dillon et al., 2010). The Eucalyptus Project is a positive example (https://www.eucalyptus.com/); it follows the approach of UDDI in the SOC/SOA paradigm, by starting the partnership with Amazon. This kind of approach will be driven forward as more partnerships with the “big players” in the business emerge (Dillon et al., 2010).
Furthermore, Wei and Blake (2010) suggest that the cloud community could learn a lesson from UDDI and the data base community. These promising examples might entice the cloud community to support federal clouds and benefit from the opportunities, which lie beyond them (Wei and Blake, 2010).
A further opportunity lies in a rapid service deployment; in fact, cloud computing is anticipated to increase their interoperability by adding new features (Wei and Blake, 2010). Additionally, as more and more public and private organizations penetrate into the cloud, this could result into a massive increase in cross-cloud deployment (Wei and Blake, 2010).
Nathuji et al. (2010) strongly suggests the introduction of a Q-Cloud Service (a cloud backup service), which “is designed to provide assurances that the performance experienced by applications is independent of whether it is consolidated with other workloads” (Nathuji et al., 2010). Q-Clouds ease performance interference by 31% and the system utilization is improved by 35% (Nathuji et al., 2010). Improvements in the cloud community like the Q-Cloud Service support the Service-Oriented Computing in a Cloud Computing environment, to benefit from synergy effects between them. This synergy can help to build a powerful architecture for the enterprise agility (Figure 5).
The integration of standards of ontologies in Web services could improve services and start “the way for automated composition and seamless interoperation” (Yu et al., 2008). Wei and Blake (2010) claim that cloud services may support this by deriving ontological information from the information (in form of data), which is stored in the cloud. Additionally, an agent-mediated ontology generation from co-located information makes it possible that Web services have computer readable languages and work like the metadata schema – with better descriptive capacity (Yu et al., 2008).
An opportunity to increase the security level may be achieved with individual or community solutions.
Eighth Mediterranean Conference on Information Systems, Verona 2014 8
Kryvinska et al. / SWOT Analysis for Soft-Computing
Figure 5. SOA, SOC and Cloud for Enterprise Agility (source MSDN, 2013).
For example, using Amazon’s EC2 requires the consumer to fill out forms and list addresses, which will be checked by Amazon and only authorized lists are in the cloud. Furthermore Amazon collaborates with Spamhaus, an anti-spam organization, to gain a high security level (Chen et al., 2010) or Alliances have been built (Cloud Security Alliance Guidance, 2011).
As availability of the cloud is limited, this challenge may hinder a widespread adoption of cloud computing (Wei and Blake, 2010).
Furthermore, due to the constant merging and relocating of companies in the IT industry, seamless interoperation of web services is a key challenge (Yu et al., 2008), which tends to be difficult to overcome.
In addition, cloud consumers still have trust issues (reasonable or not) and providers risk that they may not acquire sufficient clients.
While the trust issues for cloud environment predominantly focus on data privacy, SOC security issues are based on messaging; messages could be intercepted or inferred by the competition or anybody who can use this information (Wei and Blake, 2010). The security challenge is generally a very crucial topic for IT. This makes it also very important for the two paradigms of cloud computing and SOC and might be the biggest threat, as malware, SPAM, or spy-software are improving constantly as well.
Moreover, Tsai et al. (2010) point to the phenomenon of “security isolation”. Even if a provider may provide a safe cloud, this cloud may also be isolated from others, as there may be conflicts with other providers and interferences.
Eighth Mediterranean Conference on Information Systems, Verona 2014 9
Kryvinska et al. / SWOT Analysis for Soft-Computing
The “lose coupling” of services in SOC to a workflow, often contains more than one service provider. The issue is that most operational services are from one firm. This creates high dependence or the service providers need to manage the applications (build out of these services) on their own to keep the workflow constant (Wei and Blake, 2010).
In addition, as Tsai et al. (2011) point out, “... to provide valid and stable services for cloud users” is a significant threat for providers’ business success.
4 SOC and Cloud Combination Amazon Services
This section explores an example of SOC/SOA/Cloud usage by Amazon. We look at different Amazon Web Services (AWS) of Amazon (SaaS, PaaS and IaaS) and its partner network (APN). Amazon is seen as a big player in Web services and sets new standards with its network (Amazon Web Service, 2013). The big amount of information from Amazon and the problems to gain practical example as a member of the Amazon Partner Network (APN) suggests further research. Thus, this section demonstrates the benefits of the combination of SOC in a cloud-computing environment on the example of AWS.
Besides the vast offer of Amazon Web services the Amazon Partner Network (APN) offers huge benefits. In this Network are currently over 600 different technology partners like Adobe, IBM, SAP, and many more. If companies like to join the APN they must submit an application with standardized input fields, where prospects provide information helping Amazon including them in their Network (AWS Partner Network, 2013). When accepted in the Network the partner is subjected to the standards of the APN. This supports the interoperability and the cross-cloud infrastructure as Amazon has a registry (Amazon S3) of APIs with standard operations (Amazon Simple Storage Service, 2006).
Every independent software/service vendor (ISV) at AWS certifies himself to support its customers. Amazon guarantees that every software development works in the APN. These developments need to be submitted to Amazon otherwise it will not work in the network. This guarantee of Amazon, tackles the problem of the need of high longer-standing workflows. The software development does not need to be tested against multiple platform configurations by the provider (AWS: White Paper, 2010). Amazon will take care of this, as it promises with its service level agreement. This might fulfill the claim of Tsai et al. (2010) of a dynamic solution as the ISV develop their own and new business needs and submit it to Amazon that configures it. Of course this will take its time, but is a good approach to shrink the obstacle of the service level agreement and support the dynamic business development (Tsai et al., 2010). To fight the problem of redundancy with a new storage option called Reduced Redundancy Storage of the Amazon S3 registry, which stores data in multiple places to overcome the problem of a single server blackout. Data is saved on several servers to guarantee 99.99% availability (Amazon Simple Storage Services, 2006).
Amazon is labeled with different third party security certificates. For example the SOC 1 or SOC 2 reviews, according to SSAE 16 and ISAE 3402 standards. Furthermore, AWS received the ISO 27001 certificate and was confirmed for the data security standards DSS and PCI. Additionally, Amazon has experience in data security, thanks to years of experience in its own product selling, but if a customer
Eighth Mediterranean Conference on Information Systems, Verona 2014 10
Kryvinska et al. / SWOT Analysis for Soft-Computing
or partner still has security issues he can encrypt its data or software additionally (AWS: Risk and Compliance, 2012).
5 Conclusions
“Today, many organizations strive to cope with rapid market changes, such as evolving customer requirements and new business processes” (Wei and Blake, 2010).
In this paper, we have deliberated the prospects and encounters of Soft-Computing Paradigms and their interworking. In an SWOT analysis, we evaluated the current situation of a combination of SOA and cloud computing (C-SOA). Despite many strengths of this concept, the analysis revealed many challenges that are yet to be overcome. Still, the main opportunities to overcome these challenges seem too lie in the establishment of standards, agreements, and rules. As companies have to cope with rapid market changes, we are convinced that this situation will drive the establishment of standards.
For an illustration, we have also presented a case of C-SOA based on the real-world application of Amazon Services. We could also show that Amazon Web Services benefit from some opportunities of SOC in a cloud-computing environment as it has managed to integrate standards and structure. Still, note that this paper does not claim that the example of Amazon is the only or best solution for the future; rather we hope that this illustration contributes to eliminating a lot of the concerns of this concept.
Future work should emphasize on the current weaknesses of C-SOA and opportunities to overcome these. Particular research and development has to be put into the establishment of standards among service providers.
References
Amazon Simple Storage Services (2006). Amazon S3. Amazon Simple Storage Services,
http://docs.amazonwebservices.com/AmazonS3/latest/API/Welcome.html, Retrieved on 2014-03-21.
Amazon Web Services (2013). AWS Technology Partner. Amazon Web Services, https://aws.amazon.com/solution-providers/isv, Retrieved on 2014-03-21.
Andrews, C. (2009). SWOT: The Evolution Of IT Service Providers To Business Technology Competitors: The Business Technology Offerings Of Accenture. Capgemini, Cognizant, and IBM. Forrester Research, Cambridge, Massachusetts.
Arbab F. (2012). Will the Real Service Oriented Computing Please Stand Up? In Proceedings of the 7th international conference on Formal Aspects of Component Software (FACS 2010), Guimarães, Portugal, October 14-16, 2010, LNCS-6921, 2012, pp. 277-285.
Armbrust, M., Fox, A., Griffith, R., Joseph, A.D., Katz, R., Konwinski, A., Lee, G., Patterson, D., Rabkin, A., Stoica, I., and Zaharia, M. (2010). A view of Cloud Computing. Communications of the ACM, 53(4), 50-58.
Auer, L., Kryvinska, N., Strauss, C. and Belov E. (2011). Software-based Business Applications/Tools to Assess Complex SOA Investments - The Cross-Vendor Comparative Analysis. In Proceedings of 5th International Workshop on Engineering Complex Distributed Systems (ECDS-2011), in conjunction with CISIS-2011, June 30th-July 2nd, 2011, Seoul, Korea, pp. 397-401.
AWS Partner Network (2013). Become an APN Technology Partner. AWS Partner Network, https://aws.amazon.com/solution-providers/isv/new, Retrieved on 2014-03-21.
AWS Risk and Compliance (2012). Amazon Web Service Risk and Compliance. AWS Risk and Compliance, http://media.amazonwebservices.com/AWS_Risk_and_Compliance_Whitepaper.pdf, Retrieved on 2014-02-20.
Eighth Mediterranean Conference on Information Systems, Verona 2014 11
Kryvinska et al. / SWOT Analysis for Soft-Computing
AWS White Paper (2010). Software as a Service on AWS. AWS White Paper,
http://d36cz9buwru1tt.cloudfront.net/SaaS_whitepaper.pdf, Retrieved on 2014-03-19.
Barry, D. K. and Dick, D. (2013). Web Services, Service-Oriented Architectures, and Cloud Computing: The Savvy Manager's Guide. 2nd Edition. Morgan Kaufmann Publishers, http://www.service-architecture.com/index.html, Retrieved on 2014-03-20.
Bauer, C., Dohmen, P., and Strauss, C. (2011). Interactive Digital Signage: an Innovative Service and its Future Strategies. . In Proceedings of the First International Workshop on Frontiers in Service Transformations and Innovations (FSTI-2011), in conjunction with EIDWT 2011, IEEE Computer Society Digital Library, 2011, 137-142.
Bauer, C., Strauss, C., Stummer, C., and Trieb, A. (2013). Context-aware services in cooperative value chains: a key player-centered approach. Scalable Computing: Practice and Experience, 14(3), 139-153.
Bernroider, E. (2002). Factors in SWOT Analysis Applied to Micro, Small-to-Medium, and Large Software Enterprises: an Austrian Study. European Management Journal, 20(5), 562-573.
Brooks, F. (1987). No Silver Bullet: Essence and Accidents of Software Engineering. Computer, 20(4), 10-19.
Chellappa, R. (1997). Intermediaries in cloud-computing: A new computing paradigm. In Proceedings of INFORMS Intermediaries in Electronic Markets.
Chen, Y., Paxson, V., and Katz, R. H. (2010). What’s new about cloud computing security? University of California, Berkeley Report No. UCB/EECS-2010-5 January.
Chollet, S., Lestideau, V., Maurel, Y., Gandrille, E., Lalanda P. and Raynaud, O. (2012). Practical Use of Formal Concept Analysis in Service-Oriented Computing. In Proceedings of the 10th International Conference on Formal Concept Analysis (ICFCA 2012), Leuven, Belgium, May 7-10, 2012, LNCS-7278, pp 61-76.
Cloud Security Alliance Guidance (2011). Version 3.0. 2011.
https://cloudsecurityalliance.org/research/security-guidance/, Retrieved on 2014-02-05.
Crasso, M., Zunino, A. and Campo, M. (2011). Research Review - A Survey of Approaches to Web Service Discovery in Service-Oriented Architectures. Journal of Database Management, 22(1), 102-132.
Dillon, T., Wu, C., and Chang, E. (2010). Cloud Computing: Issues and Challenges. In Proceedings of 24th IEEE International Conference on Advanced Information Networking and Applications, pp. 27-33.
Dogac, A., Laleci, G.B., and Cingli, Y. (2002). Exploiting web service semantics: Taxonomies vs. ontologies. IEEE Data Engineering Bullettin, 25(4), 10-16.
Erdogmus, H. (2009). Cloud computing: Does nirvana hide behind the nebula? Software, IEEE, 26(2), 4-6.
Erenkrantz, J. R. (2004). Web Services: SOAP, UDDI, and Semantic Web. Institute for Software Research, Report UCI-ISR-04-3.
Fantechi, A., Gnesi, S., Lapadula, A., Mazzanti, F., Pugliese, R., and Tiezzi, F. (2012). A Logical Verification Methodology for Service-Oriented Computing. ACM Transactions on Software Engineering and Methodology, 21(3):16.
Gellersen, H-W., and Gaedke, M. (1999). Object-Oriented Web Application Development. IEEE Internet Computing, 3(1), 60-68.
Hesseldahl, A. (2012). Amazon’s Cloud is down again, taking heroku with it. http://allthingsd.com/20121022/amazons-cloud-is-down-again-taking-heroku-and-github-with-it/, Retrieved on 2014-02-05.
Hill, T. and Westbrook, R. (1997). SWOT analysis: It's time for a product recall. Long Range Planning, 30(1), 46-52.
Huang, G. Q. and Mak, K. L. (2003). Internet Applications in Product Design and Manufacturing. Heidelberg, Springer-Verlag.
Eighth Mediterranean Conference on Information Systems, Verona 2014 12
Kryvinska et al. / SWOT Analysis for Soft-Computing
Huhns, M.N., and Singh, M.P. (2005). Service-Oriented Computing: Key Concepts and Principles. IEEE Internet Computing, 7(1), 75-81.
Kryvinska, N., Auer, L. and Strauss, C. (2008). Managing an Increased Service Heterogeneity in Converged Enterprise Infrastructure with SOA. International Journal of Web and Grid Services (IJWGS), 4(4), 440 – 460.
Kryvinska, N., Auer, L. and Strauss, C. (2011a). An Approach to Extract the Business Value from SOA Services. In Proceedings of Second International Conference on Exploring Services Science (IESS 1.1), 16-18 February, 2001, Geneva, Switzerland, Springer LNBIP-82, pp. 42–52.
Kryvinska, N., Strauss, C. and Auer, L. (2011b). Synergy of two Computational Intelligence Paradigms: SOA and Mobility for the Efficient Management of Resource-Intensive Applications on Constrained Devices. Chapter 14, book Ed.: N. Bessis and F. Xhafa, (SCI-352), Springer-Verlag Berlin Heidelberg, pp. 357-382.
MSDN - Microsoft Developer Network (2013) http://msdn.microsoft.com/en-us/architecture/aa699439.aspx, Retrieved on 2014-03-25.
Mladenow, A., Kryvinska, N. and Strauss, C. (2012). Towards Cloud-Centric Service Environments. The Society of Service Science, Journal of Service Science Research, Springer, 4(2), 213-234.
Nathuji, R., Kansal, A., and Ghaffarkhah, A. (2010). Q-clouds: managing performance interference effects for qos-aware clouds. In Proceedings of the 5th European conference on Computer systems (ACM), pp. 237-250.
National Institute of Standards and Technology US (2011). Definition of Cloud Computing,
http://csrc.nist.gov/publications/nistpubs/800-145/SP800-145.pdf, Retrieved on 2014-02-20. OpenNebula.org (2012). About the OpenNebula.org Project, http://opennebula.org/about:about,
Retrieved on 2014-02-05.
Papazoglou, M. P., Traverso, P., Dustdar, S., Leymann, F., and Krämer, B. J. (2006). Service-Oriented Computing: A research roadmap. http://drops.dagstuhl.de/opus/volltexte/2006/524/pdf/05462.SWM.Paper.524.pdf, Retrieved on 2014-02-07.
Raines G. (2009). Cloud Computing and SOA. http://www.mitre.org/work/ tech_papers/tech_papers_09/09_0743/09_0743.pdf, Retrieved on 2014-02-07.
Rine, D.C., and Bhargava, B. (1992). Guest Editor’s Introduction: Object-Oriented Computing. Computer, 25(10), 6-10.
Rouse, M. (2011). Definition: Federated cloud. http://whatis.techtarget.com/definition/federated-cloud-cloud-federation, Retrieved on 2014-02-07.
Shaheen R. (2012) Guarding Against Cloud Silos With C-SOA. Cloud Industry Insights, July 24th, 2012, http://www.rackspace.com/blog/guarding-against-cloud-silos-with-c-soa/, Retrieved on 2014-04-02.
Strauss, C., Stummer, C., Bauer, C., and Trieb, A. (2009). A Networked Ubiquitous Computing Environment for Damage Prevention: A Decision Support Framework for the Insurance Sector. In Proceedings of 38th International Conference on Parallel Processing (ICPP-2009), International Workshop on Design, Optimization and Management of Heterogeneous Networked Systems (DOM-HetNetS'09), Vienna, Austria, September, 22-25 2009, 276-281.
Sun, Y., Harmer, T., and Stewart, A. (2012). Mapping High-Level Application Requirements onto
Low-Level Cloud Resources. Journal of Software Engineering and Applications, 5(11), 894-902. Tsai, W. T., Chen, Y., Bitter, G., and Miron, D. (2006). Introduction to Service-Oriented Computing,
Report, Arizona State University.
Tsai, W. T., Sun, X. and Balasooriya, J. (2010). Service-Oriented Cloud Computing Architecture. In Proceedings of Information Technology: New Generations (ITNG), 2010 Seventh International Conference on, 12-14 April 2010, pp.684-689.
Tsai. W.T., Zhong P., Balasooriya J., Yinong Chen, Xiaoying Bai, Elston J. (2011). An approach for service composition and testing for cloud computing. In Proceedings of Tenth International
Eighth Mediterranean Conference on Information Systems, Verona 2014 13
Kryvinska et al. / SWOT Analysis for Soft-Computing
Symposium on Autonomous Decentralized Systems (ISADS 2011), 23-27 March 2011, Tokyo & Hiroshima, pp. 631-636.
Wei, Y., and Blake, M. (2010) Service-oriented computing and cloud computing: Challenges and opportunities. IEEE Internet Computing, 14(6), 72-75.
Weihrich, H. (1982). The TOWS matrix: A tool for situational analysis. Long Range Planning, 15(2), 54-66.
Yu, Q., Liu, X., Bouguettaya, A., and Medjahed, B. (2008). Deploying and managing Web services: issues, solutions, and directions. The VLDB Journal, 17(3), 537-572.
Zeller, M., Grossman, R., Lingenfelder, C., Berthold, M., Marcade, E., Pechter, R., Hoskins, M., and Holada R. (2009). Open standards and cloud computing. KDD-2009 panel report, Paris, pp. 11-18.
Eighth Mediterranean Conference on Information Systems, Verona 2014 14
E-LETTER on Systems, Control, and Signal Processing
Issue 345
May 2017
Editor: Jianghai Hu
School of Electrical and Computer Engineering
Purdue University
465 Northwestern Ave
West Lafayette, IN 47907
Tel: +1 (765) 496-2395
Fax: +1 (765) 494-3371
Welcome to the 345 issue of the Eletter, available electronically here.
To submit new articles, go “Article Submissions” on the Eletter website
To unsubscribe, please send an email with the subject line “Eletter Unsubscribe”.
The next Eletter will be mailed out at the beginning of May 2017.
Contents
1. IEEE CSS Headlines
1.1 IEEE CSS Video Clip Contest: Open for Submissions
1.2 IEEE CSS Call for Nominations for 2017 Awards
1.3 IEEE CSS Technically Cosponsored Conferences
1.4 IEEE CSS Publications Content Digest
1.5 IEEE Transactions on Control Systems Technology
2. Summer Schools
2.1 SIDRA PhD Summer School in Bertinoro - Italy
2.2 American Summer School on Model Predictive Control
3. Books
3.1 Feedback Stabilization of Controlled Dynamical Systems: In Honor of Laurent Praly
4. Journals
4.1 Contents: Automatica
4.2 Contents: Evolution Equations and Control Theory
4.3 Contents: International Journal of Applied Mathematics and Computer Science
4.4 Contents: Asian Journal of Control
4.5 Contents: Control Engineering Practice
4.6 Contents: Systems & Control Letters
4.7 Contents: Nonlinear Analysis: Hybrid Systems
4.8 Contents: Mechatronics
4.9 Contents: Mechatronics
4.10 Contents: Engineering Applications of Artificial Intelligence
4.11 Contents: Journal of Process Control
4.12 Contents: Journal of Process Control
4.13 Contents: IEEE/CAA Journal of Automatica Sinica
4.14 CFP: IEEE Transactions on Control Systems Technology
4.15 CFP: Journal of Control Science and Engineering
5. Conferences
5.1 IEEE Global Conference on Signal & Information Processing
5.2 IFAC Workshop on Lagrangian and Hamiltonian Methods for Non Linear Control
5.3 Annual Allerton Conference on Communication, Control, and Computing
5.4 IEEE Ecuador Technical Chapters Meeting
5.5 Conference on Sustainable Internet and ICT for Sustainability
5.6 International Conference on Control, Automation and Systems
5.7 Workshop on Networks and Control at University of Cambridge
5.8 IFAC World Congress Workshop: ”Rigidity Theory for Multi-agent Systems Meets Parallel
Robots: Towards the Discovery of Common Models and Methods”
6. Positions
6.1 PhD: University of Suttgart, Germany
6.2 PhD: Universit´e Laval, Canada
6.3 PhD: University of Agder, Norway
6.4 PostDoc: University of California at San Diego, USA
6.5 PostDoc: CNRS – CentraleSup´elec – Univ. Paris-Sud – Univ. Paris-Saclay, France
6.6 PostDoc: Universidad T´ecnica Federico Santa Mar´ıa, Chile
6.7 PostDoc: Sandia National Laboratories, USA
6.8 PostDoc: Israel Institute of Technology
6.9 PostDoc: University of Florida, USA
6.10 PostDoc: National Institute of Informatics, Japan
6.11 PostDoc: Queen Mary University of London, UK
6.12 PostDoc: CNRS, France
6.13 PostDoc: UT-Dallas, USA
6.14 Research Fellow/Associate: National University of Singapore, Singapore
6.15 Faculty: Sharif University of Technology, Iran
6.16 Faculty: University of Louisiana at Lafayette, USA
6.17 Faculty: Zhejiang University of Technology, China
6.18 Faculty: University of Newcastle, Australia
1. IEEE CSS Headlines
1.1. IEEE CSS Video Clip Contest: Open for Submissions
Contributed by: Magnus Egerstedt, magnus@gatechedu
The 2017 IEEE Control Systems Society Video Clip Contest is now open for business!
Submissions are now accepted for the 3rd IEEE CSS Video Clip Contest (see http://www.ieeecss.org/video-contest). The purpose of this competition is to promote control theory and automatic control to a broader audience through compelling short video clips. The videos could for example focus on a particular topic or on the field in general, with the only constraint being that the video promotes the field in a visually compelling and effective manner.
The schedule for the video clip contest is:
April 15 - Open for submission
July 1 - Deadline for submitting videos
July 15 - Winners are announced
Instructions for submitting the videos and eligibility information is available at the Video Clip Contest
website:
http://www.ieeecss.org/video-contest.
All videos are judged by a jury consisting of IEEE CSS researchers, and the best three videos will receive prizes for contributing to the contest: The 1st, 2nd place, and 3rd places are awarded $1000, $500, and $250, respectively. Moreover, the 1st place winner is invited to participate in the 2017 IEEE Conference on Control Technology and Applications (http://ccta2017.ieeecss.org) on Kohala Coast, Hawaii. The winner, or the Team Leader of the winning team, will be awarded one free conference registration for the CCTA 2017 as well as reimbursement for reasonable travel expenses - to be coordinated with the Video Clip Contest Chair in advance. The best video clips will be presented to the public during an award ceremony at the CCTA 2017.
Looking forward to seeing your video clips!
Magnus Egerstedt (Contest Chair)
Angela Schoellig (Jury Chair)
Back to the contents
1.2. IEEE CSS Call for Nominations for 2017 Awards Contributed by: Joao Hespanha, hespanha@ece.ucsb.edu
IEEE Control Systems Society Call for Nominations for 2017 Awards
Nominations will open April 15 and are due by May 15, for the following IEEE Control Systems Society Awards (see http://www.ieeecss.org/awards for full details).
- George S. Axelby Outstanding Paper Award (for a paper published in 2015 or 2016 in the IEEE Transac¬tions on Automatic Control);
- IEEE Transactions on Control System Technology Outstanding Paper Award (for a paper published in 2015 or 2016 in the IEEE Transactions on Control System Technology);
- IEEE Control Systems Magazine Outstanding Paper Award (for an article published in 2015 or 2016 in the IEEE Control Systems Magazine);
- IEEE Transactions on Control of Network Systems Outstanding Paper Award (for a paper published in 2015 or 2016 in the IEEE Transactions on Control of Network Systems)
- IEEE Control Systems Technology Award (for outstanding individual or team contributions to control systems technology);
- Control Systems Society Transition to Practice Award (for a distinguished contributor to the transition of control and systems theory to practice);
- Antonio Ruberti Outstanding Young Researcher Prize (for a young researcher for innovation and impact on systems and control).
- IEEE Control Systems Society Award for Excellence in Aerospace Control (for a team or individual contribution to Aerospace Control in the previous 36 months)
The IEEE Control Systems Society strongly encourages its members to speak up and reach out to colleagues to initiate award nominations. Each year, many highly qualified individuals, teams, and papers are overlooked for nominations simply because colleagues assumed that a nomination was already being prepared by someone else on the individual’s, team’s or authors’ behalf. You may be surprised to find out that your colleagues would be very pleased to nominate you, if they had just been encouraged to do so.
Back to the contents
1.3. IEEE CSS Technically Cosponsored Conferences
Contributed by: Luca Zaccarian, CSS AE Conferences, zaccarian@laas.fr
The following conferences have been recently included in the list of events technically cosponsored by the IEEE Control Systems Society:
- XXVI International Conference on Information, Communication and Automation Technologies (ICAT 2017). Sarajevo, Bosnia and Herzegovina. Oct 26 - Oct 28, 2017. http://icat.etf.unsa.ba/
- 21st International Conference on System Theory, Control and Computing (ICSTCC 2017). Sinaia, Roma¬nia. Oct 19 - 21, 2017. http://www.icstcc2017.ac.tuiasi.ro/
- 6th International Conference on Systems and Control (ICSC 2017). Batna, Algeria. May 7 - May 9, 2017. http://lias.labo.univ-poitiers.fr/icsc/icsc2017/
- 2017 International Conference on Unmanned Aircraft Systems (ICUAS’17). Miami (FL), United States. Jun 13 - Jun 16, 2017. http://www.uasconferences.com/
For a full listing of CSS technically cosponsored conferences, please visit
http://ieeecss.org/conferences/technically-cosponsored,
and for a list of the upcoming and past CSS main conferences please visit
http://ieeecss.org/conferences
Back to the contents
1.4. IEEE CSS Publications Content Digest
Contributed by: Elizabeth Kovacs, ekovacs2@nd.edu
CSS Publications Content Digest The IEEE Control Systems Society Publications Content Digest is a novel and convenient guide that helps readers keep track of the latest published articles. The CSS Publications Content Digest, available at http://ieeecss.org/publications-content-digest provides lists of current tables of contents of the periodicals sponsored by the Control Systems Society.
Each issue offers readers a rapid means to survey and access the latest peer-reviewed papers of the IEEE
Control Systems Society. We also include links to the Society’s sponsored Conferences to give readers a
preview of upcoming meetings.
Back to the contents
1.5. IEEE Transactions on Control Systems Technology Contributed by: Michelle Colasanti, ieeetcst@osu.edu
Table of Contents
IEEE Transactions on Control Systems Technology
Volume 25 (2017), Issue 3 (May)
Regular papers
- Data and Reliability Characterization Strategy for Automatic Detection of Valve Stiction in Control Loops,
O. Pozo Garcia, A. Zakharov, and S.-L. J¨ams¨a-Jounela, page 769
- Plug-and-Play Voltage Stabilization in Inverter-Interfaced Microgrids via a Robust Control Strategy, M.
S. Sadabadi, Q. Shafiee, and A. Karimi, page 781
- Identification and Multivariable Gain-Scheduling Control for Cloud Computing Systems, P. S. Saikrishna,
R. Pasumarthy, and N. P. Bhatt, page 792
- Optimum Adaptive Piecewise Linearization: An Estimation Approach in Wind Power, M. Vaezi, P.
Khayyer, and A. Izadian, page 808
- Overshoot-Free Steering-Based Source Seeking, A. Raisch and M. Krstic, page 818
- Resource-Efficient Gradient Methods for Model Predictive Pulse Pattern Control on an FPGA, S. Richter,
T. Geyer, and M. Morari, page 828
- Critical-to-Fault-Degradation Variable Analysis and Direction Extraction for Online Fault Prognostic, C. Zhao and F. Gao, page 842
- Improving Demand Response Potential of a Supermarket Refrigeration System: A Food Temperature
Estimation Approach, R. Pedersen, J. Schwensen, B. Biegel, T. Green, and J. Stoustrup, page 855 - Bilateral Control of SeNZA—A Series Hybrid Electric Bicycle, M. Corno, F. Roselli, and S. M. Savaresi,
page 864
- Speed Advisory and Signal Offsets Control for Arterial Bandwidth Maximization and Energy Consumption Reduction, G. De Nunzio, G. Gomes, C. Canudas-de-Wit, R. Horowitz, and P. Moulin, page 875
- An Alternative Approach to the Inference of the Extended Observability Matrix, and Its Relation With the PO-MOESP Algorithm, R. Nava Cabrera, V. M. Alvarado, G. Lopez Lopez, and M. Adam Medina, page 888
- Distributed Model Predictive Control for Heterogeneous Vehicle Platoons Under Unidirectional Topologies, Y. Zheng, S. E. Li, K. Li, F. Borrelli, and J. K. Hedrick, page 899
- Health-Aware and User-Involved Battery Charging Management for Electric Vehicles: Linear Quadratic Strategies, H. Fang, Y. Wang, and J. Chen, page 911
- JLS-PPC: A Jump Linear System Framework for Networked Control, B. L. Reed and F. S. Hover, page 924
- A Hybrid Dynamic System Model for Multimodal Transportation Electrification, A. M. Farid, page 940 - Organic Rankine Cycle for Vehicles: Control Design and Experimental Results, J. Peralez, M. Nadri, P. Dufour, P. Tona, and A. Sciarretta, page 952
- Robust Real-Time Needle Tracking in 2-D Ultrasound Images Using Statistical Filtering, K. Mathiassen,
D. Dall’Alba, R. Muradore, P. Fiorini, and O. J. Elle, page 966
- Control Engineering Methods for the Design of Robust Behavioral Treatments, K. Bekiroglu, C. Lagoa,
S. A. Murphy, and S. T. Lanza, page 979
- A Passivity-Based Approach for Stable Patient–Robot Interaction in Haptics-Enabled Rehabilitation Sys
tems: Modulated Time-Domain Passivity Control, S. F. Atashzar, M. Shahbazi, M. Tavakoli, and R. V.
Patel, page 991
- Force Feedback Control Assisted Tympanostomy Tube Insertion, W. Liang and K. K. Tan, page 1007
Brief papers
- Flocking for Multirobots Without Distinguishing Robots and Obstacles, D. Sakai, H. Fukushima, and F.
Matsuno, page 1019
- System Identification and Fault Diagnosis of an Electromagnetic Actuator, A. Forrai, page 1028
- Modeling and Vibration Control for a Moving Beam With Application in a Drilling Riser, W. He, S. Nie,
T. Meng, and Y.-J. Liu, page 1036
- Online Learning-Based Server Provisioning for Electricity Cost Reduction in Data Center, J. Yang, S.
Zhang, X. Wu, Y. Ran, and H. Xi, page 1044
- Toward a Robust Motorcycle Braking, M. E.-H. Dabladji, D. Ichalal, H. Arioui, and S. Mammar, page
1052
- Allocating Sensors and Actuators via Optimal Estimation and Control, J. A. Taylor, N. Luangsomboon,
and D. Fooladivanda, page 1060
- Robust Control Allocation for Spacecraft Attitude Tracking Under Actuator Faults, Q. Shen, D. Wang,
S. Zhu, and E. K. Poh, page 1068
- State Estimation of Macromotion Positioning Tables Based on Switching Kalman Filter, Y. Li, Y. Tan,
R. Dong, and H. Li, page 1076
- An Adaptive Two-Level Quantizer for Networked Control Systems, D. Almakhles, A. K. Swain, A. Nasiri,
and N. Patel, page 1084
- Variational Bayesian Gaussian Mixture Regression for Soft Sensing Key Variables in Non-Gaussian Indus
trial Processes, J. Zhu, Z. Ge, and Z. Song, page 1092
- Policy Iteration Approach to Control Residual Gas Fraction in IC Engines Under the Framework of
Stochastic Logical Dynamics, Y. Wu and T. Shen, page 1100
- Subspace Identification of 2-D CRSD Roesser Models With Deterministic-Stochastic Inputs: A State
Computation Approach, A. Alenany, G. Merc`ere, and J. A. Ramos, page 1108
- 3-D Trajectory Planning of Aerial Vehicles Using RRT*, P. Pharpatara, B. H´eriss´e, and Y. Bestaoui, page
1116
- A Probabilistic Just-in-Time Learning Framework for Soft Sensor Development With Missing Data, X.
Yuan, Z. Ge, B. Huang, and Z. Song, page 1124
- Modular-Controller-Design-Based Fast Terminal Sliding Mode for Articulated Exoskeleton Systems, T.
Madani, B. Daachi, and K. Djouani, page 1133
- General Unbiased FIR Filter With Applications to GPS-Based Steering of Oscillator Frequency, Y. S.
Shmaliy, S. H. Khan, S. Zhao, and O. Ibarra-Manzano, page 1141
CALL FOR PAPERS
- Special Issue on System Identification and Control in Biomedical Applications, page 1152
Back to the contents
2. Summer Schools
2.1. SIDRA PhD Summer School in Bertinoro - Italy
Contributed by: M. Elena Valcher, meme@dei.unipd.it
2017 PhD Summer School in Bertinoro - Italy
The SIDRA PhD Summer School is a one-week annual event organized by SIDRA (Societ`a Italiana Docenti e Ricercatori in Automatica), the Italian Control Systems Society, that takes place in Bertinoro, Forl`ı-Cesena, Italy.
The summer school is primarily intended for, but not restricted to, PhD, post-graduate, and researcher candidates in the Control Systems and Automatica fields. Students coming from other disciplines may also benefit from the school depending on the topics discussed.
The school success encountered in these recent years and the distinguished international guests, involved a progressive internationalization of the school that now opens up to receive applications from all European PhD students.
The two themes presented this year are “Formal Methods for the Control of Large-scale Networked Nonlinear Systems with Logic Specifications“, coordinated by professors Maria Domenica Di Benedetto and Giordano Pola (University of L’Aquila), and “Port-Hamiltonian modelling and passivity-based control of physical systems. Theory and applications“, coordinated by professors Alessandro Macchelli (University of Bologna) and Cristian Secchi (University of Modena and Reggio).
Click on the School web site http://sidra2017.dei.unibo.it/to have information about the program, the venue, the accommodation, the instructions for reaching the school, the instructions for applying to the school and the requested fee.
General informations
• Location: as usual, the SIDRA school will take place in Bertinoro, FC Italy.
• Period: from Monday, July 3rd, 2017 to Saturday, July 8th, 2017.
• Accommodation: the attendees will be hosted in the Centro Universitario di Bertinoro. The accommoda-tions will be assigned according to the booking date. Refer to Mrs. Monica Michelacci for any question.
• Costs: the fee includes the registration to the School and accommodation (half-board, i.e. breakfast and
lunch). The fee (arrival on July 2nd, departure on July 8th) is:
- 500,00 Euro (in double room)
- 630,00 Euro (in single room)
You could indicate your roommate in the NOTE field of the registration form.
Although non-encouraged, it is possible to register for only one of the two modules of the School; in this
case, the registration fee is 330 Euro (double room) or 430 Euro (single room).
• Registration deadlines:
- June 5, 2017: Fill in the form before this deadline to submit your application to the school.
- June 12, 2017: You will receive a acceptance/rejection communication by email and instructions for the
final registration.
- June 23, 2017: deadline for final registration, with the payment of the fee.
• Certified Final Exam: it will be possible to take a certified exam at the end of the school. In case of positive result, the certification can be used by students to obtain the corresponding ECTS from their PhD courses, or to comply with any other requests by their home university.
Contacts:
For questions related to payment, accommodation and logistics:
Mrs. Monica Michelacci mmichelacci@ceub.it
For questions related to the School and the courses:
Prof. Claudio Melchiorri mailto:claudio.melchiorri@unibo.it
Prof. Maria Elena Valcher meme@dei.unipd.it
Link web site: http://sidra2017.dei.unibo.it/
Back to the contents
2.2. American Summer School on Model Predictive Control
Contributed by: Sasa V. Rakovic, sasa.v.rakovic@gmail.com
This is an update in regards to the final programme of, and registration for, the American MPC summer school 2017.
First American Summer School on Model Predictive Control (MPC)
July 25, 2017 – July 28, 2017
University of Wisconsin-Madison
http://mpc-summer-school.che.wisc.edu/
Registration opens May 1, 2017, closes May 14, 2017 or when the class is full
The first American summer school on MPC will be held at the University of Wisconsin-Madison from July 25, 2017 to July 28, 2017. The summer school is organized by Sasa V. Rakovic, James B. Rawlings and Ilya V. Kolmanovsky and is supported in part by the National Science Foundation.
The summer school will enable up to 50 participants (university graduate students as well as interested re-searchers and control practitioners) from a cross-cutting set of disciplines in engineering, science, and applied mathematics to receive advanced education and training in the theory, implementation and applications of MPC. The instructors are international experts and leading researchers with diversity of backgrounds and disciplines in engineering, science and applied mathematics.
The main topics of the summer school are:
- Introduction to MPC and MPC essentials (by William S. Levine).
- Classical MPC: regulation, estimation, and disturbance models (by James B. Rawlings).
- Robust MPC (by Sasa V. Rakovic).
- Stochastic MPC (by Ilya V. Kolmanovsky).
- Economic MPC (by David Angeli).
- Online optimization for MPC (by Lorentz T. Biegler).
- Industrial applications of MPC (by Thomas A. Badgwell).
The summer school will provide a carefully crafted overview of the theoretical fundamentals of MPC and state-of-the-art numerical methods and software for implementing the advanced MPC methods on difficult and challenging examples and industrial applications. The summer school will also feature mini-projects and discussions that will enable all attendees to present, and discuss, problems of direct interest to their research/professional work, and also to receive feedback from a set of instructors with valuable expertise in all areas of MPC research.
Registration:
The registration is a two-stage process. The first stage consists of online registration at http://mpc-summer-school.che.wisc.edu/. The online registration opens on May 01, 2017. Those interested in attending the summer school should register at their earliest convenience. The registration form will require (1) your name, (2) contact information (email, address and telephone), (3) the name of your organization, (4) briefly state your motivation for taking the course in terms of its relevance to your research/professional work and (5) if you are applying for a travel stipend. Because of the anticipated interest and space limitations, we will notify participants about the admission decisions starting May 15, 2017 on an individual basis. If admitted, a participant will be provided with a link to pay with a credit card the registration fee of $250 within one week.
Travel and accommodation support:
We anticipate being able to offer a limited number of travel stipends to graduate students from American universities attending the summer school to partially offset their travel costs. The admitted participants who apply for a stipend will be notified about the travel stipend decision and the amount that will be reimbursed. The stipend will be paid after the completion of the summer school. In terms of accommodation arrangements, an on-campus housing option at about $25 per night will be made available.
Additional information:
Further information about the summer school will be made available at
http://mpc-summer-school.che.wisc.edu/
Back to the contents
3. Books
3.1. Feedback Stabilization of Controlled Dynamical Systems: In Honor of Laurent Praly Contributed by: Yasmin Brookes, yasmin.brookes@springer.com
Feedback Stabilization of Controlled Dynamical Systems: In Honor of Laurent Praly
by Nicolas Petit
ISBN: 978-3-319-51297-6
March 2017, Springer
Paperback, 354 pages, $159.00/euro 119,99
http://www.springer.com/gb/book/9783319512976
This book is a tribute to Professor Laurent Praly and follows on from a workshop celebrating the occasion of his 60th birthday.
It presents new and unified visions of the numerous problems that Laurent Praly has worked on in his prolific career: adaptive control, output feedback and observers, stability and stabilization. His main contributions are the central topic of this book.
The book collects contributions written by prominent international experts in the control community, ad¬dressing a rich variety of topics: emerging ideas, advanced applications, and theoretical concepts. Organized in three sections, the first section covers the field of adaptive control, where Laurent Praly started his career. The second section focuses on stabilization and output feedback, which is also the topic of the second half of his career. Lastly, the third section presents the emerging research that will form Laurent Praly’s scientific legacy.
Contents
Part I Adaptive Control
1 Lyapunov Functions Obtained from First Order Approximations
2 A Review on Model Reduction by Moment Matching for Nonlinear Systems
3 Event-Triggered Control of Nonlinear Systems: A Small-Gain Approach
Part II Stabilization and Output Feedback
4 An ODE Observer for Lyapunov-Based Global Stabilization of a Bioreactor Nonlinear PDE
5 From Pure State and Input Constraints to Mixed Constraints in Nonlinear Systems
6 Output Regulation via Low-Power Construction
7 Passivity-Based Control of Mechanical Systems
8 Asymptotic Stabilization of Some Finite and Infinite Dimensional Systems by Means of Dynamic Event Triggered Output Feedbacks
9 Incremental Graphical Asymptotic Stability for Hybrid Dynamical Systems
Part III New Perspectives
10 Exponential Stability of Semi-linear One-Dimensional Balance Laws
11 Checkable Conditions for Contraction After Small Transients in Time and Amplitude
12 Asymptotic Expansions of Laplace Integrals for Quantum State Tomography
13 Recent Developments in Stability Theory for Stochastic Hybrid Inclusions
Back to the contents
4. Journals
4.1. Contents: Automatica
Contributed by: Elisa Capello, elisa.capello@polito.it
Table of Contents
Automatica
Vol. 79, May 2017
http://www.sciencedirect.com/science/journal/00051098/79
- Augusto Ferrante, Alexander Lanzon, Lorenzo Ntogramatzidis, “Discrete-time negative imaginary sys
tems”, pages 1-10.
- Vladimir L. Kharitonov, “Prediction-based control for systems with state and several input delays”, pages
11-16.
- Wei Guo, Zhi-Chao Shao, Miroslav Krstic, “Adaptive rejection of harmonic disturbance anticollocated
with control in 1D wave equation”, pages 17-26.
- Qingling Zhang, Li Li, Xing-Gang Yan, Sarah K. Spurgeon, “Sliding mode control for singular stochastic
Markovian jump systems with uncertainties”, pages 27-34.
- Benjamin Noack, Joris Sijs, Marc Reinhardt, Uwe D. Hanebeck, “Decentralized data fusion with inverse
covariance intersection”, pages 35-41.
- Shuai Feng, Pietro Tesi, “Resilient control under Denial-of-Service: Robust design”, pages 42-51.
- Alexey S. Matveev, Anna A. Semakova, Andrey V. Savkin, “Tight circumnavigation of multiple moving
targets based on a new method of tracking environmental boundaries”, pages 52-60.
- Alan Tapia, Miguel Bernal, Leonid Fridman, “An LMI approach for second-order sliding set design using
piecewise Lyapunov functions”, pages 61-64.
- Song Fang, Jie Chen, Hideaki Ishii, “Design constraints and limits of networked feedback in disturbance attenuation: An information-theoretic analysis”, pages 65-77.
- Wei Zhu, Huizhu Pu, Dandan Wang, Huaqing Li, “Event-based consensus of second-order multi-agent systems with discrete time”, pages 78-83.
- Wei Liu, Jie Huang, “Adaptive leader-following consensus for a class of higher-order nonlinear multi-agent systems with directed switching networks”, pages 84-92.
- Ran Huang, Jinhui Zhang, Zhongwei Lin, “Decentralized adaptive controller design for large-scale power systems”, pages 93-100.
- Mengyuan Fang, Yucai Zhu, “Analysis of over-sampling based identification”, pages 101-107.
- Chandrashekar Lakshminarayanan, Shalabh Bhatnagar, “A stability criterion for two timescale stochastic approximation schemes”, pages 108-114.
- Sofie Haesaert, Paul M.J. Van den Hof, Alessandro Abate, “Data-driven and model-based verification via Bayesian identification and reachability analysis”, pages 115-126.
- Lassi Paunonen, David Seifert, “Asymptotic behaviour in the robot rendezvous problem”, pages 127-130. - Torsten S¨oderstr¨om, Umberto Soverini, “Errors-in-variables identification using maximum likelihood esti-mation in the frequency domain”, pages 131-143.
- Patrizio Tomei, “Multi-sinusoidal disturbance rejection for discrete-time uncertain stable systems”, pages 144-151.
- Xunyuan Yin, Jinfeng Liu, “Distributed moving horizon state estimation of two-time-scale nonlinear sys-tems”, pages 152-161.
- Hamid Maarouf, “The resolution of the equation image and the pole assignment problem: A general approach”, pages 162-166.
- Deyuan Meng, Kevin L. Moore, “Convergence of iterative learning control for SISO nonrepetitive systems subject to iteration-dependent uncertainties”, pages 167-177.
- Konstantin Zimenko, Denis Efimov, Andrey Polyakov, Wilfrid Perruquetti, “A note on delay robustness for homogeneous systems with negative degree”, pages 178-184.
- Henrik Anfinsen, Mamadou Diagne, Ole Morten Aamo, Miroslav Krstic, “Estimation of boundary param¬eters in general heterodirectional linear hyperbolic systems”, pages 185-197.
- Zhongyang Fei, Shuang Shi, Chang Zhao, Ligang Wu, “Asynchronous control for 2-D switched systems with mode-dependent average dwell time”, pages 198-206.
- Yanqiong Zhang, Zhenhua Deng, Yiguang Hong, “Distributed optimal coordination for multiple hetero-geneous Euler–Lagrangian systems”, pages 207-213.
- Amit Prakash Pandey, Maur´ıcio C. de Oliveira, “A new discrete-time stabilizability condition for Linear Parameter-Varying systems”, pages 214-217.
- Orest V. Iftime, “On the Newton–Kleinman method for strongly stabilizable infinite-dimensional systems”, pages 218-222.
- G. Roberto Marseglia, Davide M. Raimondo, “Active fault diagnosis: A multi-parametric approach”, pages 223-230.
- Petro Feketa, Humberto Stein Shiromoto, Sergey Dashkovskiy, “Almost ISS property for feedback con¬nected systems”, pages 231-234.
- Bilal Gunes, Jan-Willem van Wingerden, Michel Verhaegen, “Predictor-Based Tensor Regression (PBTR) for LPV subspace identification”, pages 235-243.
- Rong Su, Bengt Lennartson, “Control protocol synthesis for multi-agent systems with similar actions
instantiated from agent and requirement templates”, pages 244-255.
- Jiarao Huang, Dawei Shi, Tongwen Chen, “Energy-based event-triggered state estimation for hidden
Markov models”, pages 256-264.
- Shuyou Yu, Ting Qu, Fang Xu, Hong Chen, Yunfeng Hu, “Stability of finite horizon model predictive
control with incremental input constraints”, pages 265-272.
- Lo¨ıc Bourdin, Emmanuel Tr´elat, “Linear–quadratic optimal sampled-data control problems: Convergence
result and Riccati theory”, pages 273-281.
- Deyuan Meng, “Bipartite containment tracking of signed networks”, pages 282-289.
- Xianghua Wang, Chee Pin Tan, Donghua Zhou, “A novel sliding mode observer for state and fault esti
mation in systems not satisfying matching and minimum phase conditions”, pages 290-295.
- Florian D¨orfler, Sergio Grammatico, “Gather-and-broadcast frequency control in power systems”, pages
296-305.
- Matthias A. M¨uller, “Nonlinear moving horizon estimation in the presence of bounded disturbances”,
pages 306-314.
- Lilian K. Carvalho, Marcos V. Moreira, Jo˜ao Carlos Basilio, “Diagnosability of intermittent sensor faults in discrete event systems”, pages 315-325
- Giulia Prando, Alessandro Chiuso, Gianluigi Pillonetto, “Maximum Entropy vector kernels for MIMO system identification”, pages 326-339.
- Giordano Scarciotti, Alessandro Astolfi,“Data-driven model reduction by moment matching for linear and nonlinear systems”, pages 340-351.
EDITORIALS
- Andrew R. Teel, “An abrupt, yet smooth, transition”, page 352.
- Andrew R. Teel, “Transition in an editorship”, page 353.
Back to the contents
4.2. Contents: Evolution Equations and Control Theory
Contributed by: Irena Lasiecka, lasiecka@memphis.edu
Content of Evolution Equations and Control Theory, Vol 6, Nr 2, 2017.
http://aimsciences.org/journals/contentsListnew.jsp?pubID=946
1. Stability of ground states for logarithmic Schr¨odinger equation with a S'-interaction Pages : 155 - 175, Alex H. Ardila
2. Asymptotic for the perturbed heavy ball system with vanishing damping term Pages : 177 - 186, Mounir Balti and Ramzi May
3. Optimal control for a hyperbolic problem in composites with imperfect interface: A memory effect Pages : 187 - 217, Luisa Faella and Carmen Perugia
4. Existence of periodic solution for a Cahn-Hilliard/Allen-Cahn equation in two space dimensions Pages : 219 - 237, Changchun Liu and Hui Tang
5. General decay for a viscoelastic Kirchhoff equation with Balakrishnan-Taylor damping, dynamic boundary conditions and a time-varying delay term Pages : 239 - 260, Wenjun Liu, Biqing Zhu, Gang Li and Danhua Wang
6. Viscoelastic plate equation with boundary feedback Pages : 261 - 276, Muhammad I. Mustafa
7. Periodic solutions for time-dependent subdifferential evolution inclusions Pages : 277 - 297, Nikolaos S. Papageorgiou and Vicent¸iu D. R˘adulescu
8. A note on dimension reduction for unbounded integrals with periodic microstructure via the unfolding
method for slender domains Pages : 299 - 318, Elvira Zappale
Back to the contents
4.3. Contents: International Journal of Applied Mathematics and Computer Science Contributed by: AMCS, amcs@uz.zgora.pl
International Journal of Applied Mathematics and Computer Science (AMCS)
2017, Volume 27, Number 1 (March)
Regular issue
www.amcs.uz.zgora.pl
CONTENTS
- Luis-Delgado J.D., Al-Hadithi B.M. and Jim´enez A. A novel method for the design of switching surfaces
for discretized MIMO nonlinear systems 5
- Moysis L. and Karampetakis N.P. Construction of algebraic and difference equations with a prescribed
solution space 19
- Sajewski L. Minimum energy control of descriptor fractional discrete-time linear systems with two different
fractional orders 33
- Xu F., Puig V., Ocampo-Martinez C., Olaru S. and Niculescu S.-I. Robust MPC for actuator-fault tolerance
using set-based passive fault detection and active fault isolation 43
- Tomera M. Hybrid switching controller design for the maneuvering and transit of a training ship 63
- Rodr´ıguez-Li˜n´an M.C., Mendoza M., Bonilla I. and Ch´avez-Olivares C.A. Saturating stiffness control of
robot manipulators with bounded inputs 79
- Kruthika H.A.,Mahindrakar A.D. and Pasumarthy R. Stability analysis of nonlinear time-delayed systems
with application to biological models 91
- Tenne Y. Machine-learning in optimization of expensive black-box functions 105
- Brugno A., D’Apice C., Dudin A. and Manzo R. Analysis of an MAP/PH/1 queue with flexible group
service 119
- Widuch J. A relation of dominance for the bicriterion bus routing problem 133
- Liu Y., Qin K., Rao C. and Alhaji Mahamadu M. Object-parameter approaches to predicting unknown
data in an incomplete fuzzy soft set 157
- Szemenyei M. and Vajda F. Dimension reduction for objects composed of vector sets 169
- Xue Y., Liu P., Tao Y. and Tang X. Abnormal prediction of dense crowd videos by a purpose-driven lattice
Boltzmann model 181
- Ptak R., ˙Zygad lo B. and Unold O. Projection-based text line segmentation with a variable threshold 195
- Kubica M. and Kania D. Area-oriented technology mapping for LUT-based logic blocks 207
Back to the contents
4.4. Contents: Asian Journal of Control
Contributed by: Lichen Fu, lichen@ntu.edu.tw
Asian Journal of Control Vol.19, No.2 March, 2017 CONTENTS
[Regular Paper]
1. A Simultaneous Mixed LQR/H Control Approach to the Design of Reliable Active Suspension Con-trollers (pages 415–427), Author: Jenq-Lang Wu
2. L2–Optimal Fopdt Models of High–Order Transfer Functions (pages 428–437), Authors: Daniele Casagrande, Wieslaw Krajewski and Umberto Viaro
3. Hysteresis-Based Design of Dynamic Reference Trajectories to Avoid Saturation in Controlled Wind Tur-bines (pages 438–449), Christian Tutiv´en, Yolanda Vidal, Leonardo Acho and Jos´e Rodellar
4. Improved control performance of the 3-DoF aeroelastic wing section: a TP model based 2D parametric control performance optimization (pages 450–466), Alexandra Szollosi and Peter Baranyi
5. More Relaxed Non-Quadratic Stabilization Conditions Using Ts Open Loop System and Control Law Properties (pages 467–481), Navid Vafamand and Mokhtar Shasadeghi
6. Alignment Motion Control for an Automated Human Ear Surgery via Vision-Servoing (pages 482–493), Wenchao Gao, Wenyu Liang and Kok Kiong Tan
7. Stability of Local On-Ramp Metering Control Laws (pages 494–509), Luis Alvarez-Icaza, Oscar Rosas-Jaimes and Mar´ıa Elena L´arraga
8. Synchronization of General Linear Multi-Agent Systems With Measurement Noises (pages 510–520), Wenhui Liu, Chunjie Yang, Feiqi Deng and Jiarong Liang
9. Fractional Order Modeling And Nonlinear Fractional Order Pi-Type Control For PMLSM System (pages 521–531), Bao Song, Shiqi Zheng, Xiaoqi Tang and Wenjun Qiao
10. Observability and Controllability Analysis for Micro-Positioning Stage Described by Sandwich Model with Hysteresis (pages 532–542), Na Luo, Yonghong Tan and Ruili Dong
11. A Delay-Dependent Approach to Robust Fast Adaptive Fault Estimation Design for Uncertain Neutral Systems with Time-Varying Interval Delay (pages 543–553), Fuqiang You, Hui Li, Fuli Wang and Shouping Guan
12. An Efficient Finite Difference Method for The Time-Delay Optimal Control Problems With Time-Varying Delay (pages 554–563), Amin Jajarmi and Mojtaba Hajipour
13. Event-Based Semiglobal Consensus of Homogenous Linear Multi-Agent Systems Subject to Input Satu-ration (pages 564–574), Bo Zhou, Xiaofeng Liao, Tingwen Huang, Huaqing Li and Guo Chen
14. Robust Finite-Time H Control of a Class of Disturbed Systems using Lmi-Based Approach (pages 575–586), Xiaoyu Zhang, Jihong Zhong, Quan Zhang and Kemao Ma
15. Smith Predictor Based Fractional-Order-Filter PID Controllers Design for Long Time Delay Systems (pages 587–598), Maamar Bettayeb, Rachid Mansouri, Ubaid Al-Saggaf and Ibrahim Mustafa Mehedi
16. Output Feedback Control of Surge and Rotating Stall in Axial Compressors (pages 599–605), Hanlin Sheng, Wei Huang and Tianhong Zhang
17. Crossed Synchronization of Multiple Subnets Complex Network System with Time-Varying Delay (pages 606–613), Zhou Bi-feng, Lou Yi-ping and Zhong Yao-xiang
18. Distributed Consensus of Multi-Agent Networks Via Event-Triggered Pinning Control (pages 614–624), Dan Liu, Aihua Hu and Dan Zhao
19. Robust Output Feedback Controller Design for Time-Delayed Teleoperation: Experimental Results (pages 625–635), I. Sharifi, H. A. Talebi and M. Motaharifar
20. Fault Diagnosis and Sliding Mode Fault Tolerant Control for Non-Gaussian Stochastic Distribution Control Systems Using T-S Fuzzy Model (pages 636–646), Yao Lina and Lei Chunhui
21. Performance Analysis of The Auxiliary-Model-Based Multi-Innovation Stochastic Newton Recursive Al-gorithm for Dual-Rate Systems (pages 647–658), Pengfei Cao and Xionglin Luo
22. Non-Fragile Observer-Based H Control for Uncertain Neutral-Type Systems via Sliding Mode Tech
nique (pages 659–671), Zhen Liu, Cunchen Gao, Huimin Xiao and Yonggui Kao
23. L2-gain analysis and anti-windup design of switched linear systems subject to input saturation (pages 672–680) Author: Xinquan Zhang
24. Event-Triggered Control for Couple-Group Multi-Agent Systems with Logarithmic Quantizers and Com-munication Delays (pages 681–691), Mei Yu, Chuan Yan and Dongmei Xie
25. Modeling and Control Approach to Coupled Tanks Liquid Level System Based on Function-Type Weight RBF-ARX Model (pages 692–707), Feng Zhou, Hui Peng, Xiaoyong Zeng, Xiaoying Tian and Jun Wu
26. Delay-Dependent Stability Criterion for Discrete-Time Systems with Time-Varying Delays (pages 708–716), Changchun Hua, Shuangshuang Wu, Zhenhua Bai and Xinping Guan
27. A Matrix Approach to the Analysis and Control of Networked Evolutionary Games with Bankruptcy Mechanism (pages 717–727), Shihua Fu, Yuzhen Wang and Guodong Zhao
28. Modal Kalman Filter (pages 728–738), Gh. Mohammaddadi, N. Pariz and A. Karimpour
29. New Upper Matrix Bounds with Power Form for the Solution of the Continuous Coupled Algebraic Riccati Matrix Equation (pages 739–747), Jianzhou Liu, Yanpei Wang and Juan Zhang
30. Vector-Based Adaptive Attitude Observer and Controller on Special Orthogonal Group (pages 748–764), Xuhui Lu, Yingmin Jia and Fumitoshi Matsuno
31. Fault-Tolerant Finite Frequency H Control for Uncertain Mechanical System with Input Delay and
Constraint (pages 765–780), Shidong Xu, Guanghui Sun and Weichao Sun
[Brief Paper]
1. Iterative Path Integral Approach to Nonlinear Stochastic Optimal Control Under Compound Poisson Noise (pages 781–786), Okumura Yuta, Kenji Kashima and Yoshito Ohta
2. Exponential Stability for Multi-Area Power Systems with Time Delays Under Load Frequency Controller Failures (pages 787–791), Xu Li, Rui Wang, Shu-Nan Wu and Georgi M. Dimirovski
3. Homogeneous Control of Pneumatic Cylinders Based on Time Delay Model and Artstein Transformation (pages 792–798), E. Edjekouane, S. Riachy, M. Ghanes and J.-P. Barbot
4. A Robust Fault Estimation Scheme for a Class of Nonlinear Systems (pages 799–804), W. S. Chua, C. P. Tan, M. Aldeen and S. Saha
5. Algebraic Connectivity Estimation Based on Decentralized Inverse Power Iteration (pages 805–812), Yue Wei, Hao Fang, Jie Chen and Bin Xin
6. Particle Smoother for Nonlinear Systems With One-Step Randomly Delayed Measurements (pages
813–819), Huang Yu-Long and Zhang Yong-Gang
Back to the contents
4.5. Contents: Control Engineering Practice
Contributed by: Martin B¨ock, cep@acin.tuwien.ac.at
Control Engineering Practice
Volume 62
May 2017
- Hongquan Ji, Xiao He, Jun Shang, Donghua Zhou, Incipient fault detection with smoothing techniques in
statistical process monitoring,Pages 11-21
- Wilber Acu˜na-Bravo, Enrico Canuto, Marco Agostani, Marco Bonadei, Proportional electro-hydraulic
valves: An Embedded Model Control solution,Pages 22-35
- Berk Altin, Jeroen Willems, Tom Oomen, Kira Barton, Iterative Learning Control of Iteration-Varying
Systems via Robust Update Laws with Experimental Implementation,Pages 36-45
- Yiqi Liu, Yongping Pan, Daoping Huang, Qilin Wang, Fault prognosis of filamentous sludge bulking using
an enhanced multi-output gaussian processes regression,Pages 46-54
- Carlos Rossa, Mahdi Tavakoli, Issues in closed-loop needle steering,Pages 55-69
- Ping Shen, Han-Xiong Li, The consistency control of mold level in casting process,Pages 70-78
- Xi Ma, Jinqiu Hu, Laibin Zhang, EMD-based online Filtering of Process Data,Pages 79-91
- Li-Juan Li, Ting-Ting Dong, Shu Zhang, Xiao-Xiao Zhang, Shi-Ping Yang, Time-delay identification in
dynamic processes with disturbance via correlation analysis,Pages 92-101
- I. Miletovic, D.M. Pool, O. Stroosma, M.M. van Paassen, Q.P. Chu, Improved Stewart platform state
estimation using inertial and actuator position measurements,Pages 102-115
- Zhiyuan Tang, David J. Hill, Tao Liu, Two-stage voltage control of subtransmission networks with high
penetration of wind power,Pages 1-10
Back to the contents
4.6. Contents: Systems & Control Letters
Contributed by: John Coca, j.coca@elsevier.com
Systems & Control Letters
Volume 102
April 2017
- Khaled Bahlali, Meriem Mezerdi, Brahim Mezerdi, Existence and optimality conditions for relaxed mean
field stochastic control problems, Pages 1-8
- Myung-Gon Yoon, Consensus of adaptive multi-agent systems, Pages 9-14
- Matthew C. Turner, Positive modification as an anti-windup mechanism, Pages 15-21
- Lin Zhao, Yingmin Jia, Jinpeng Yu, Adaptive finite-time bipartite consensus for second-order multi-agent
systems with antagonistic interactions, Pages 22-31
- Johan Markdahl, Jens Hoppe, Lin Wang, Xiaoming Hu, A geodesic feedback law to decouple the full and
reduced attitude, Pages 32-41
- Christian Commault, Jacob van der Woude, Taha Boukhobza, On the fixed controllable subspace in linear
structured systems, Pages 42-47
- P. Pepe, I. Karafyllis, Z.P. Jiang, Lyapunov–Krasovskii characterization of the input-to-state stability for
neutral systems in Hale’s form, Pages 48-56
- Brandon J. Wellman, Jesse B. Hoagg, A flocking algorithm with individual agent destinations and without
a centralized leader, Pages 57-67
- Delsin Menolascino, ShiNung Ching, Bispectral analysis for measuring energy-orientation tradeoffs in the
control of linear systems, Pages 68-73
- Kaihong Yang, Haibo Ji, Hierarchical analysis of large-scale control systems via vector simulation function,
Pages 74-80
- Fengwei Chen, Marion Gilson, Hugues Garnier, Tao Liu, Robust time-domain output error method for
identifying continuous-time systems with time delay, Pages 81-92
- Walid Djema, Fr´ed´eric Mazenc, Catherine Bonnet, Stability analysis and robustness results for a nonlinear
system with distributed delays describing hematopoiesis, Pages 93-101
- Quanxin Zhu, Feng Jiang, Hui Wang, Bao Wang, Comment on “Stability analysis of stochastic differential
equations with Markovian switching” [Systems & Control Letters 61 (2012) 1209–1214], Pages 102-103
- Yijing Xie, Zongli Lin, Global optimal consensus for multi-agent systems with bounded controls, Pages
104-111
- Qingbin Gao, Nejat Olgac, Stability analysis for LTI systems with multiple time delays using the bounds
of its imaginary spectra, Pages 112-118
Back to the contents
4.7. Contents: Nonlinear Analysis: Hybrid Systems
Contributed by: John Coca, j.coca@elsevier.com
Nonlinear Analysis: Hybrid Systems
Volume 24
May 2017
- Qiumei Zhang, Daqing Jiang, Yanan Zhao, Donal O’Regan, Asymptotic behavior of a stochastic population
model with Allee effect by L´evy jumps, Pages 1-12
- Yujie Zhang, Yongsheng Ou, Xinyu Wu, Yimin Zhou, Resilient dissipative dynamic output feedback control
for uncertain Markov jump Lur’e systems with time-varying delays, Pages 13-27
- R. Rakkiyappan, V. Preethi Latha, Quanxin Zhu, Zhangsong Yao, Exponential synchronization of Marko
vian jumping chaotic neural networks with sampled-data and saturating actuators, Pages 28-44
- Le Van Hien, Hieu Trinh, Switching design for suboptimal guaranteed cost control of 2-D nonlinear switched
systems in the Roesser model, Pages 45-57
- Yazhou Tian, Yuanli Cai, Yuangong Sun, Stability of switched nonlinear time-delay systems with stable
and unstable subsystems, Pages 58-68
- Seung Woo Lee, Sung Jin Yoo, Adaptive-observer-based output-constrained tracking of a class of arbitrarily
switched uncertain non-affine nonlinear systems, Pages 69-82
- Behrooz Rahmani, Robust output feedback sliding mode control for uncertain discrete time systems, Pages
83-99
- W.P.M.H. Heemels, V. Sessa, F. Vasca, M.K. Camlibel, Computation of periodic solutions in maximal
monotone dynamical systems with guaranteed consistency, Pages 100-114
- Bing Cui, Chunhui Zhao, Tiedong Ma, Chi Feng, Leaderless and leader-following consensus of multi-agent
chaotic systems with unknown time delays and switching topologies, Pages 115-131
- Tae H. Lee, Ju H. Park, Improved criteria for sampled-data synchronization of chaotic Lur’e systems using
two new approaches, Pages 132-145
- Lijuan Zha, Jian-an Fang, Xiaofan Li, Jinliang Liu, Event-triggered output feedback control for networked
Markovian jump systems with quantizations, Pages 146-158
- Xueling Li, Xiangze Lin, Shihua Li, Yun Zou, Globally smooth output feedback stabilization of a class of
planar switched systems with average dwell time, Pages 159-170
- Mengling Li, Feiqi Deng, Almost sure stability with general decay rate of neutral stochastic delayed hybrid
systems with L´evy noise, Pages 171-185
- Min Meng, James Lam, Jun-e Feng, Xudong Zhao, Xiaoming Chen, Exponential stability analysis and
synthesis of positive T-S fuzzy systems with time-varying delays, Pages 186-197
- Corentin Briat, Dwell-time stability and stabilization conditions for linear positive impulsive and switched
systems, Pages 198-226
- Ding Zhai, An-Yang Lu, Dan Ye, Qing-Ling Zhang, Adaptive tracking control for a class of switched un
certain nonlinear systems under a new state-dependent switching law, Pages 227-243
- Zidong Ai, Cancan Chen, Asymptotic stability analysis and design of nonlinear impulsive control systems,
Pages 244-252
Back to the contents
4.8. Contents: Mechatronics
Contributed by: John Coca, j.coca@elsevier.com
Mechatronics
Vol. 42
April 2017
- Wenyu Liang, Wenchao Gao, Kok Kiong Tan, Stabilization system on an office-based ear surgical device by force and vision feedback, Mechatronics, Pages 1-10
- David Rijlaarsdam, Pieter Nuij, Johan Schoukens, Maarten Steinbuch, A comparative overview of frequency domain methods for nonlinear systems, Pages 11-24
- M.A. Oliver-Salazar, D. Szwedowicz-Wasik, A. Blanco-Ortega, F. Aguilar-Acevedo, R. Ruiz-Gonz´alez, Characterization of pneumatic muscles and their use for the position control of a mechatronic finger, Pages 25-40
- Sang Hun Woo, Sung Mok Kim, Min Gun Kim, Byung-Ju Yi, Wheekuk Kim, Torque-balancing algorithm for the redundantly actuated parallel mechanism, Mechatronics, Pages 41-51
- Michael Muehlebach, Raffaello D’Andrea, The Flying Platform – A testbed for ducted fan actuation and control design, Pages 52-68
- Lei Zhou, Mohammad Imani Nejad, David L. Trumper, One-axis hysteresis motor driven magnetically suspended reaction sphere, Pages 69-80
- Li Chen, Fengyu Liu, Jian Yao, Zhao Ding, Chunhao Lee, Chi-kuan Kao, Farzad Samie, Ying Huang, Chengliang Yin, Design and validation of clutch-to-clutch shift actuator using dual-wedge mechanism, Pages 81-95
Back to the contents
4.9. Contents: Mechatronics
Contributed by: John Coca, j.coca@elsevier.com
Mechatronics
Vol. 43
May 2017
- Sergey Edward Lyshevski, Control of high-precision direct-drive mechatronic servos: Tracking control with
adaptive friction estimation and compensation, Pages 1-5
- Stijn Derammelaere, Bram Vervisch, Jasper De Viaene, Kurt Stockman, Sensorless load angle control for
two-phase hybrid stepper motors, Pages 6-17
- A. Gonzalez-Rodriguez, F.J. Castillo-Garcia, E. Ottaviano, P. Rea, A.G. Gonzalez-Rodriguez, On the ef
fects of the design of cable-Driven robots on kinematics and dynamics models accuracy, Pages 18-27
- Ehsan Hashemi, Saeid Khosravani, Amir Khajepour, Alireza Kasaiezadeh, Shih-Ken Chen, Bakhtiar Litk
ouhi, Longitudinal vehicle state estimation using nonlinear and parameter-varying observers, Pages 28-39
- Wiput Tuvayanond, Manukid Parnichkun, Position control of a pneumatic surgical robot using PSO based
2-DOF loop shaping structured controller, Pages 40-55
- Yijie Guo, Sandipan Mishra, Constrained optimal iterative learning control with mixed-norm cost func
tions, Pages 56-65
- Jin-hui Fang, Fan Guo, Zheng Chen, Jian-hua Wei, Improved sliding-mode control for servo-solenoid valve
with novel switching surface under acceleration and jerk constraints, Pages 66-75
- Chang-Hyuk Lee, Jiwon Choi, Hooman Lee, Joongbae Kim, Kyung-min Lee, Young-bong Bang, Exoskele
tal master device for dual arm robot teaching, Pages 76-85
- Tiemin Li, Fengchun Li, Yao Jiang, Jinglei Zhang, Haitong Wang, Kinematic calibration of a 3-P(Pa)S
parallel-type spindle head considering the thermal error, Pages 86-98
- Alessandro Beghi, Fabio Marcuzzi, Paolo Martin, Fabio Tinazzi, Mauro Zigliotto, Virtual prototyping of
embedded control software in mechatronic systems: A case study, Pages 99-111
- Sunkyum Yoo, Jaeyoul Lee, Jaeyeon Choi, Goobong Chung, Wan Kyun Chung, Development of rotary
hydro-elastic actuator with robust internal-loop-compensator-based torque control and cross-parallel con
nection spring, Pages 112-123
Back to the contents
4.10. Contents: Engineering Applications of Artificial Intelligence
Contributed by: John Coca, j.coca@elsevier.com
Engineering Applications of Artificial Intelligence
Volume 61
May 2017
- XiaoHong Han, Long Quan, XiaoYan Xiong, Matt Almeter, Jie Xiang, Yuan Lan, A novel data clustering algorithm based on modified gravitational search algorithm, Pages 1-7
- Ahmad M. El-Nagar, Mohammad El-Bardini, Parallel realization for self-tuning interval type-2 fuzzy con-troller, Pages 8-20
- Geovanny Osorio, Thibaud Monteiro, Lorraine Trilling, Fr´ed´eric Albert, Multi-criteria assignment policies to improve global effectiveness of medico-social service sector, Pages 21-34
- Mehdi Akbari, Hassan Rashidi, Sasan H. Alizadeh, An enhanced genetic algorithm with new operators for task scheduling in heterogeneous computing systems, Pages 35-46
- Sukjun Lee, David Enke, Youngmin Kim, A relative value trading system based on a correlation and rough set analysis for the foreign exchange futures market, Pages 47-56
- Fangwei Zhang, Jianbo Li, Jihong Chen, Jing Sun, Augustine Attey, Hesitant distance set on hesitant fuzzy sets and its application in urban road traffic state identification, Pages 57-64
- Sy Dzung Nguyen, Huu-Vinh Ho, Thoi-Trung Nguyen, Nang Toan Truong, Tae-Il Seo, Novel fuzzy sliding controller for MRD suspensions subjected to uncertainty and disturbance, Pages 65-76
- Arijit De, Sri Krishna Kumar, Angappa Gunasekaran, Manoj Kumar Tiwari, Sustainable maritime inven¬tory routing problem with time window constraints, Pages 77-95
- Anh-Tu Nguyen, Raymundo M´arquez, Antoine Dequidt, An augmented system approach for LMI-based control design of constrained Takagi-Sugeno fuzzy systems, Pages 96-102
- R. Venkata Rao, Dhiraj P. Rai, Joze Balic, A multi-objective algorithm for optimization of modern ma¬chining processes, Pages 103-125
- Salim Zair, Sylvie Le H´egarat-Mascle, Evidential framework for robust localization using raw GNSS data, Pages 126-135
- Kevin Kam Fung Yuen, The fuzzy cognitive pairwise comparisons for ranking and grade clustering to build a recommender system: An application of smartphone recommendation, Pages 136-151
- Pablo J. Prieto, Nohe R. Cazarez-Castro, Luis T. Aguilar, Selene L. Cardenas-Maciel, Chattering existence and attenuation in fuzzy-based sliding mode control, Pages 152-160
- Wensheng Gan, Jerry Chun-Wei Lin, Philippe Fournier-Viger, Han-Chieh Chao, Jimmy Ming-Tai Wu, Justin Zhan, Extracting recent weighted-based patterns from uncertain temporal databases, Pages 161-172
Back to the contents
4.11. Contents: Journal of Process Control
Contributed by: John Coca, j.coca@elsevier.com
Journal of Process Control
Volume 52
April 2017
- John Perkins, Thomas McAvoy, Wolfgang Marquardt, Denis Dochain, History and growth of Journal of Process Control, Pages A1-A4
- Xiaoqiang Wang, Vladimir Mahalec, Feng Qian, Globally optimal nonlinear model predictive control based on multi-parametric disaggregation, Pages 1-13
- Mariusz Buciakowski, Marcin Witczak, Vicen¸c Puig, Damiano Rotondo, Fatiha Nejjari, J´ozef Korbicz, A bounded-error approach to simultaneous state and actuator fault estimation for a class of nonlinear systems, Pages 14-25
- Qian-Fang Liao, Da Sun, Wen-Jian Cai, Shao-Yuan Li, You-Yi Wang, Type-1 and Type-2 effective Takagi-Sugeno fuzzy models for decentralized control of multi-input-multi-output processes, Pages 26-44
- Young Jae Choung, Jihoon Kang, Seoung Bum Kim, Process control of time-varying systems using parameter-less self-organizing maps, Pages 45-56
- G. Lloyds Raja, Ahmad Ali, Smith predictor based parallel cascade control strategy for unstable and inte-grating processes with large time delay, Pages 57-65
- Seokho Kang, Pilsung Kang, An intelligent virtual metrology system with adaptive update for semicon¬ductor manufacturing, Pages 66-74
- Afrooz Ebadat, Patricio E. Valenzuela, Cristian R. Rojas, Bo Wahlberg, Model Predictive Control oriented experiment design for system identification: A graph theoretical approach, Pages 75-84
Back to the contents
4.12. Contents: Journal of Process Control
Contributed by: John Coca, j.coca@elsevier.com
Journal of Process Control
Volume 53
May 2017
- D.K.M. Kufoalor, G. Frison, L. Imsland, T.A. Johansen, J.B. Jørgensen, Block factorization of step response model predictive control problems, Pages 1-14
- Alireza Fatehi, Biao Huang, Kalman filtering approach to multi-rate information fusion in the presence of irregular sampling rate and variable measurement delay, Pages 15-25
- K. Vinther, Rene J. Nielsen, Palle Andersen, Jan D. Bendtsen, Optimization of interconnected absorption cycle heat pumps with micro-genetic algorithms, Pages 26-36
- C.J. Muller, I.K. Craig, Economic hybrid non-linear model predictive control of a dual circuit induced draft cooling water system, Pages 37-45
- Michalis Frangos, Uncertainty quantification for cuttings transport process monitoring while drilling by ensemble Kalman filtering, Pages 46-56
- Jaeheum Jung, Won Je Lee, Sangmin Park, Younghun Kim, Chul-Jin Lee, Chonghun Han, Improved control strategy for fixed-speed compressors in parallel system, Pages 57-69
- Long Teng, Youyi Wang, Wenjian Cai, Hua Li, Robust model predictive control of discrete nonlinear systems with time delays and disturbances via T–S fuzzy approach, Pages 70-79
- Norelys Aguila-Camacho, Johan D. Le Roux, Manuel A. Duarte-Mermoud, Marcos E. Orchard, Control of a grinding mill circuit using fractional order controllers, Pages 80-94
- Gheorghe Maria, Mara Cri¸san, Operation of a mechanically agitated semi-continuous multi-enzymatic
reactor by using the Pareto-optimal multiple front method, Pages 95-105
Back to the contents
4.13. Contents: IEEE/CAA Journal of Automatica Sinica
Contributed by: YangQuan Chen, yqchen@ieee.org
IEEE/CAA Journal of Automatica Sinica (JAS) published papers on FOSC (fractional order systems and controls) are free to download.
IEEE/CAA Journal of Automatica Sinica (JAS) is a joint publication of the IEEE and the Chinese Asso¬ciation of Automation. The objective of this journal is high quality and rapid publication of articles, with a strong focus on new trends, original theoretical and experimental research and developments, emerging technologies, and industrial standards in automation.
Special Issues on “Fractional Order Systems and Controls (FOSC)”, guest co-edited by Prof. YangQuan Chen, University of California, Merced, USA; Prof. Dingy ¨u Xue, Northeastern University, China, and Prof. Antonio Visioli, University of Brescia, Italy, have published 54 papers so far from 2015 to 2017.
It is a great pleasure to announce that, all these published FOSC papers were compiled in a single indexable
PDF file (31.8Mb)
http://mechatronics.ucmerced.edu/sites/mechatronics.ucmerced.edu/files/page/documents/ieee caa journal of automatica s
2017.pdf
to share in public domain. For individual papers, they are listed with a link to local PDF for your easy
reading, and the LaTeX BiBTeX library file) for easy citation
http://mechatronics.ucmerced.edu/sites/mechatronics.ucmerced.edu/files/page/documents/ieee caa journal of automatica s
2017.bib .txt
Visit http://mechatronics.ucmerced.edu/jas-si-fosc for the original call for papers, two editorials, and the single combined PDF file for all published papers, LaTeX BiBTeX file, and a list of all papers with a hyper-link to each paper and its local PDF.
Back to the contents
4.14. CFP: IEEE Transactions on Control Systems Technology
Contributed by: Guillaume Merc`ere, guillaume.mercere@univ-poitiers.fr
CFP: Special Issue on System identification and control in biomedical applications in IEEE Transactions on Control Systems Technology
Contributions are invited for a special issue of the IEEE Transactions on Control Systems Technology devoted to the subject of System Identification and Control in Biomedical Applications. The purpose of this special issue is to document the current status of research in this field through an original collection of diverse, high-quality papers. The emphasis is on the role control systems technology plays in advancing the state of the art in the challenges of applying feedback control in living organisms, with emphasis on biomedicine. Specifically, we aim at (i) pointing out theoretical and practical issues specific to bio-medical systems, (ii) bringing together solutions developed under different settings with specific attention to the validation of these tools in bio-medical settings using real-life datasets and experiments, and (iii) introducing significant case studies. Topics of common interests include (but are not limited to) the following:
- theoretical and implementation challenges which arise in medical systems,
- control engineering tools for solving specific system design problems in medical technology,
- novel data-driven modeling techniques capturing the dynamics of biomedical systems, and accounting for intra- and inter-individual variability,
- evidence of successful projects in biomedicine enabled by system identification and control, such as the artificial pancreas and closed-loop anesthesia.
- application areas in healthcare and medical systems, such as assistive devices and therapeutics in medical rehabilitation, and mathematical models of infectious disease spread.
- prevention and treatment of chronic, relapsing disorders and illnesses such as cancer, diabetes, obesity, and HIV.
Only contributions that include significant results based on analysis of real data or experimental validation will be included. Papers must contain high-quality original contributions and be prepared in accordance with the IEEE Transactions on Control Systems Technology standards. Prospective authors should state in their cover letter and in the notes section of the submission site that their manuscript is intended for the special issue on “system identification and control in biomedical applications.” Submitted manuscripts must not have been previously published or be under review for possible publication elsewhere.
Time line:
Manuscripts Due: November 1, 2017
Notification to authors (after the first round of reviews): March 1, 2018
Notification of final decision: June 1, 2018
Publication Date: January 2019
Authors can submit their manuscripts via https://mc.manuscriptcentral.com/tcst
Information for Authors prior to submitting a paper is available via
http://www.ieeecss.org/publications/tcst/information-authors
All inquiries should be directed to G. Merc`ere you can contact via his email address: guillaume.mercere@univ-poitiers.fr
Guest Editors:
Guillaume Mercere, Universitede Poitiers, France (LEAD)
Bayu Jayawardhana, University of Groningen, The Netherlands
Alexander Medvedev, Uppsala University, Sweden
Daniel E. Rivera, Arizona State University, Tempe, Arizona, USA
Caterina Scoglio, Kansas State University, Manhattan, Kansas, USA
Back to the contents
4.15. CFP: Journal of Control Science and Engineering
Contributed by: Tushar Jain, tushar@iitmandi.ac.in
CFP: Special Issue on System Modeling, Control, and Diagnosis for Energy Efficiency in Buildings, Journal of Control Science and Engineering (Deadline Extended)
Energy utilization in buildings relies on numerous factors, such as building structure, energy management systems design, and effective control and maintenance under the varying thermal or cooling load. Their heterogeneous operational characteristics contribute to serious environmental and economic problems due to excessive consumption of energy and other resources.
Consequently, there is a growing interest in high performance buildings where the underlying concept of performance incorporates energy efficiency, thermal performance, and healthy indoor environment.
Achieving this high performance objective is mainly dependent on enhanced control strategies and the continuous commissioning of the building Heating, Ventilation and Air-Conditioning (HVAC) systems under the constraints of economically managing the energy flows within the building to meet the needs of the occupants. The related challenges encompass describing the complex nonlinear dynamics of the building, deriving mathematical models for control, and deploying different control strategies for different weather conditions and occupancy profile.
Even when the building automation system or when advanced controllers are applied to enhance system efficiency, faults can occur during installation, routine operations, or scheduled preventive maintenances, resulting in excessive energy waste. This calls for more sophisticated and tailored algorithms for analysis and control, yielding energy efficient solutions for smart buildings.
The purpose of this special issue is primarily to publish high quality research papers as well as review articles on recent advances on operating buildings in an energy efficient way through building and HVAC modeling, diagnostics, and controls. Original contributions that are not yet published or that are not currently under review in other journals or peer-reviewed conferences are invited, in particular, manuscripts containing novel ideas and algorithms with practical/experimental applications.
Potential topics include, but are not limited to, the following:
Monozone/multizone modeling approaches and HVAC components modeling
Building simulation tools and platforms
Optimal supervisory control and model-based predictive control for building systems
Energy-optimal control for space-conditioning systems
Fault detection and diagnosis of HVAC and building systems
Fault-tolerant control of HVAC systems
Continuous commissioning
Whole-building optimization
Green energy rating systems in buildings
Economic performance analysis of the building
Authors can submit their manuscripts through the Manuscript Tracking System at http://mts.hindawi.com/submit/journals/jcse/smcd/.
Manuscript Due: Friday, 16 June 2017
First Round of Reviews: Friday, 8 September 2017
Publication Date: Friday, 3 November 2017
Lead Guest Editor:
Tushar Jain, Indian Institute of Technology Mandi, Himachal Pradesh, India
Guest Editors:
Joseph J. Yame, University of Lorraine, Nancy, France
Alessandro Beghi, University of Padova, Padova, Italy
Du Zhimin, Shanghai Jiao Tong University, Shanghai, China
Learn more about this topic at https://www.hindawi.com/journals/jcse/si/368954/cfp/
Back to the contents
5. Conferences
5.1. IEEE Global Conference on Signal & Information Processing
Contributed by: Cedric Langbort, langbort@illinois.edu
5th IEEE Global Conference on Signal & Information Processing
Symposium on Control and Information Theoretic Approaches to Privacy and Security
November 14-16, 2017, Montreal, Canada
http://www.ieeeglobalsip.org
http://www.ieeeglobalsip.org/GS17 CfP ITPS.pdf
The ubiquity of technologies such as wireless communications, biometric identification systems, on-line data repositories, and smart electricity grids, has created new challenges in information security and privacy. Traditional approaches based on cryptography are far from adequate in such complex systems and funda¬mentally new techniques must be developed. Control and Information theory provide fundamental limits that can guide the development of methods for addressing these challenges. Historically, both Systems and Control and the Information Theory communities have developed independent approaches to deal with the issue of security and privacy. But various emerging applications require tools from both theories to be used in tandem. There has been relatively little effort in bringing the two fields together and have a cohesive discussion on modeling and solution approaches to security and privacy. The symposium aims to serve as such a venue that discusses the perspectives developed by both communities in a timely and productive manner. Topics of interest include but are not limited to:
• Modeling systems under cyber and physical attacks
• Intrusion detection and attack identification
• Secure state estimation and communication
• Game theoretic, supervisory, and robust control approaches to security and privacy
• Secrecy and secret key capacity of wireless channel
• Secure communication under adversarial attack
• Practical code design for physical layer security
• Secure cross-layer design techniques
• Secure communication with an uncertain physical layer • Jamming-assisted secure wireless transmission
• Security and Privacy issues in applications (e.g., Smart Grids, UAVs, etc.)
Paper Submission:
Prospective authors are invited to submit full-length papers (up to 4 pages for technical content including figures and possible references, and with one additional optional 5th page containing only references) and extended abstracts (up to 2 pages, for paper-less industry presentations and Ongoing Work presentations) via the GlobalSIP 2017 conference website. Manuscripts should be original (not submitted/published anywhere else) and written in accordance with the standard IEEE double-column paper template. Accepted full-length papers will be indexed on IEEE Xplore. Accepted abstracts will not be indexed in IEEE Xplore, however the abstracts and/or the presentations will be included in the IEEE SPS SigPort. Accepted papers and abstracts will be scheduled in lecture and poster sessions.
Important Dates:
- May 15, 2017: Paper submission due
- June 30, 2017: Notification of Acceptance
- July 22, 2017: Camera-ready papers due
General Co-Chairs:
Aditya Mahajan, McGill University (aditja.mahajan@mcgill.ca)
Ashish Khisti, University of Toronto (akhisti@ece.utoronto.ca)
Rafael F. Schaefer, Technische Universit¨at Berlin (rafael.schaefer@tu-berlin.de)
Technical Co-Chairs:
Cedric Langbort, University of Illinois at Urbana-Champaign (langbort@illinois.edu)
Holger Boche, Technische Universit¨at M¨unchen (boche@tum.de)
Back to the contents
5.2. IFAC Workshop on Lagrangian and Hamiltonian Methods for Non Linear Control Contributed by: Juan I. Yuz, juan.yuz@usm.cl
6th IFAC Workshop on Lagrangian and Hamiltonian Methods for Non Linear Control
LHMNLC18, 2-4 May 2018, Valpara´ıso, Chile
First Announcement and Call for Papers
Hosting Institution: Universidad T´ecnica Federico Santa Mar´ıa - UTFSM, Valpara´ıso, Chile
Sponsored by: IFAC International Federation of Automatic Control, IFAC TC Non Linear Control Systems Co-sponsored by: IFAC TC Distributed Parameter Systems, IFAC TC Control Design, IEEE CSS TC on DPS
Scope: Recent technological progresses in material science, actuators, and sensors as well as in real-time computing have induced the necessity of accounting for nonlinear and distributed parameters phenomena in the design of the nonlinear control, sometimes including the design of the plant. A very efficient design method, based on the Lagrangian and Hamiltonian formulations of physical systems’ dynamics, has been increasingly developed and used in the last years. These formulations allow to combine the powerful design methods using passivity-based control with the specific properties of the differential-geometric structure of the Lagrangian and Hamiltonian systems. The application areas include robotics, tele-manipulation and power systems where developments concerning control systems interacting through a communication network are important. Recent developments have shown that generalization of the Hamiltonian and Lagrangian frameworks can be used for distributed parameter systems with applications in fluid systems and fluid-structure interactions, as well as irreversible thermodynamical systems with applications to chemical and biological processes and smart materials.
Topics: This workshop will cover new developments in modelling nonlinear distributed parameters control theory and applications that have been recently developed to take advantage of and to exploit the mathe¬matical structures common to the multi-physical systems. The workshop program will include both regular papers and posters. The format of the workshop will encourage in-depth and fruitful discussion between all the participants.
Location: The workshop will be held at Universidad T´ecnica Federico Santa Maria (www.usm.cl) in Val-paraiso, one of the most prestigious engineering universities of Latin America. Built upon dozens of steep hillsides overlooking the Pacific Ocean, Valparaiso, also known as the Jewel of the Pacific, boasts a labyrinth of graffiti filled streets and cobblestone alleyways, embodying a rich architectural and cultural legacy and hosting one of Pablo Neruda’s houses. Valparaiso’s historic quarter is an UNESCO World Heritage Site since 2003, thanks to its historical importance, natural beauty and unique architecture.
Important dates:
Submission of draft papers, invited sessions proposal, and abstracts for poster session: October 15, 2017
Author notification: January 14, 2018
Final paper due: February 28, 2018
For more information: www.lhmnlc18.org
International Program Committee (IPC) co-chairs
Yann Le Gorrec, FEMTO-ST, UBFC, France E-mail: legorrec@femto-st.fr
Martin Guay, Queen’s University, Canada, E-mail: martin.guay@chee.queensu.ca
National Organizing Committee (NOC) co-chairs
Juan I. Yuz, UTFSM, Valparaiso, Chile, E-mail: juan.yuz@usm.cl
Juan C. Ag¨uero UTFSM, Valparaiso, Chile, E-mail: juan.aguero@usm.cl
Back to the contents
5.3. Annual Allerton Conference on Communication, Control, and Computing Contributed by: Rachel Palmisano, rep2@illinois.edu
FIFTY-FIFTH ANNUAL ALLERTON CONFERENCE ON COMMUNICATION, CONTROL, AND COM
PUTING
October 3, 2017 – Opening Tutorials
October 4-6, 2017 – Conference Sessions
CALL FOR PAPERS
The Fifty-Fifth Annual Allerton Conference on Communication, Control, and Computing will kick off with Opening Tutorials being held on Tuesday, October 3, 2017 at the Coordinated Science Laboratory. The conference sessions will start on Wednesday, October 4, 2017 through Friday, October 6, 2017, at the Allerton Park and Retreat Center. The Allerton House is located twenty-six miles southwest of the Urbana-Champaign campus of the University of Illinois in a wooded area on the Sangamon River. It is part of the fifteen-hundred acre Robert Allerton Park, a complex of natural and man-made beauty designated as a National natural landmark. Allerton Park has twenty miles of well-maintained trails and a living gallery of formal gardens, studded with sculptures collected from around the world.
Papers presenting original research are solicited in the areas of:
biological information systems
coding techniques and applications
coding theory
data storage
information theory
multiuser detection and estimation
network information theory
sensor networks in communications
wireless communication systems
intrusion/anomaly detection and diagnosis
network coding
network games and algorithms
performance analysis
pricing and congestion control
reliability, security and trust
decentralized control systems
robust and nonlinear control
adaptive control and automation
robotics
distributed and large-scale systems
complex networked systems
optimization
dynamic games
machine learning and learning theory
signal models and representations
signal acquisition, coding, and retrieval
detection and estimation
learning and inference
statistical signal processing
sensor networks
data analytics.
Final versions of papers that are presented at the conference are required to be submitted electronically by October 8, 2017 in order to appear in the Conference Proceedings and IEEE Xplore.
PLENARY LECTURE is scheduled for Friday, October 6, 2017 at the Allerton Park and Retreat Center. (we will add the speaker info when confirmed)
OPENING TUTORIAL LECTURES will be presented on Tuesday, October 3, 2017 at the Coordinated Sci¬ence Laboratory, University of Illinois at Urbana-Champaign. (we will add the speakers info when confirmed)
INFORMATION FOR AUTHORS: Regular papers suitable for presentation in twenty minutes are solicited. Regular papers will be published in full (subject to a maximum length of eight 8.5” x 11” pages, in two column format) in the Conference Proceedings. Only papers that are actually presented at the conference and uploaded as final manuscripts can be included in the proceedings, which will be available after the conference on IEEE Xplore.
For reviewing purposes of papers, a title and a five to ten page extended abstract, including references and sufficient detail to permit careful reviewing, are required.
Manuscripts can be submitted during June 16-July 10, 2017 with the submission deadline of July 10th being firm. Please follow the instructions at the Conference website: http://www.csl.illinois.edu/allerton/.
Authors will be notified of acceptance via e-mail by August 7, 2017, at which time they will also be sent
detailed instructions for the preparation of their papers for the Proceedings.
Important Dates:
Submission Deadline: July 10, 2017
Acceptance Date: August 7, 2017
Registration Opens: after August 7, 2017
Conference Dates: October 3-6, 2017
Final Submission Deadline: October 8, 2017
Conference Co-Chairs: Naira Hovakimyan and Negar Kiyavash
Email: amellis@illinois.edu URL:
http://www.csl.illinois.edu/allerton/
Back to the contents
5.4. IEEE Ecuador Technical Chapters Meeting
Contributed by: Alberto Sanchez, aesanchez@ieee.org
CFP: IEEE Ecuador Technical Chapters Meeting (ETCM) 2017 http://sites.ieee.org/etcm-2017
IEEE Ecuador section takes great pleasure in inviting you to the 2017 IEEE ETCM, which will be held for the first time from October 16th-20th in Salinas, Ecuador.
The 2017 IEEE Ecuador Technical Chapters Meeting (ETCM) will be the second edition of the running series of conferences organized by the IEEE Ecuador Section and which intends to provide a highly prestigious venue for researchers, students and practitioners from the IEEE Technical Society Chapters in Ecuador.
The conference covers both theoretical and practical issues related to Communications, Computing, Control Systems, Industrial Electronics, Engineering in Medicine and Biology, Power and Energy, Robotics and Automation. Topics of interest, but not limited to, are:
SYSTEMS AND CONTROL
Adaptive Systems, Signal Processing, Embedded Systems, Fault Tolerant Systems, Identification, Predictive
control.
INDUSTRIAL ELECTRONICS
Power Converters, Power semiconductors, Machines and drives, Power electronics in transportation systems,
Power electronics applications.
COMMUNICATIONS
Internet of Things, Communications Systems Security, Green Communications, Wireless Communications, Optical Communications, Waveforms and Signal Processing, Access Networks and Systems, Cluster, Grid, P2P Cloud Computing, Satellite and Space Communications, Networking protocols and performance.
COMPUTER
Security and Privacy, Semantic Computing, Real Time Systems, Multimedia Computing, Learning Tech
nologies, Distributed Processing, Human Computer Interaction, Computer Vision, Data Engineering.
COMPUTATIONAL INTELLIGENCE
Neural Networks, Fuzzy Systems, Evolutionary and Swarm Computation, Learning Systems, Data Science.
POWER AND ENERGY
Transmission, Distribution, Power Generation, Power System Control & Operation, Reliability, Stability,
Renewables, SmartGrids.
ENGINEERING IN MEDICINE AND BIOLOGY
Clinical Engineering, Telemedicine, and Health Care, Bioinformatics, Biomechanics, Biomaterials, Bioinstru
mentation, Signal and Image Processing, Biophysics.
ROBOTICS AND AUTOMATION SYSTEMS
Automation, Automation in Logistics and Supply Chain Management, Sensors, Robotics, Assistive Technolo¬gies, System Integration, Sensor/Actuator Networks, Distributed and Cloud Robotics, Autonomous Vehicles, Human/Robot Interaction.
Important Dates
Full Paper Submission: 7 July 2017
Acceptance Notification: 31 July 2017
Final paper Submission: 15 August 2017
Workshops & Tutorials: 16-17 October 2017
Conference Dates: 18-20 October 2017
Back to the contents
5.5. Conference on Sustainable Internet and ICT for Sustainability
Contributed by: Nuno Pereira, nap@isep.ipp.pt
The 5th Conference on Sustainable Internet and ICT for Sustainability (SustainIT 2017)
Funchal, Portugal - December 6-7, 2017
https://sustainit2017.m-iti.org
CALL FOR PAPERS
The 5th Conference on Sustainable Internet and ICT for Sustainability (SustainIT 2017) will be held De¬cember 6-7, 2017 in Funchal, Portugal. Papers are invited in all aspects of Sustainable Internet and ICT, Sustainability through the application of ICT, and Human-Centered technology for sustainability, including works that report on prototype test best and real-world deployments.
Ultimately, the goal of this conference is to bring together people from different research areas, and provide a forum to exchange ideas, discuss solutions, and share experiences among researchers, professionals, and application developers from both industry and academia. The topics of interest include but are not limited to the following:
*Sustainable Internet and ICT
1. Green Internet (e.g., novel standards and metrics for green communications, measurement and evaluation of the Internet’s sustainability)
2. Energy-efficient data centers (e.g., algorithms for reduced power, energy and heat, trade-offs between energy efficiency, Quality of Service, and reliability)
3. Adaptation of computing and communications infrastructure to variable renewable energy supply
4. Emerging computing / storage technologies for energy efficient operation
5. E-waste (e.g. obsolescence of electronic equipment and its disposal issues)
*Sustainability through the application of ICT
1. ICT for energy efficiency in smart homes and buildings
2. ICT for energy efficiency in industrial environments
3. ICT for smart grids and water distribution systems
4. ICT for sustainable transport and logistics
5. ICT for monitoring and conservation of biodiversity (e.g., underwater and fauna monitoring)
*Human-Centered Technology for Sustainability
1. User evaluation of test-bed and prototype implementations
2. Metrics for sustainability and their evaluation
3. Behavior change regarding sustainability choices
4. Human-factors in sustainable ICT systems
5. Novel user interfaces and interaction techniques
PROGRAM CHAIRS
Mario Berg´es, Carnegie Mellon University, Pittsburgh, PA, USA
Lucas Pereira, Madeira Interactive Technologies Institute, Funchal, PT
IMPORTANT DATES (TIME IS 23:59 AOE)
* June 23 - title and abstract registration
* July 7 - paper submission
* September 21 - WIP / Demo / PhD Forum submission
* October 8 - notification of acceptance
* October 21 - WIP / Demo / PhD Forum acceptance notification
* November 10 - camera ready version
* December 6 and 7 - conference
TYPES OF SUBMISSIONS (IEEE COMPUTER SOCIETY TEMPLATE)
* Papers - up to 10 pages (8 + 2 additional)
* WIP - up to 4 pages (3 + 1 additional)
* Demo - up to 2 pages
* PhD Forum - up to 3 pages
(additional pages are subject to an extra charge)
Back to the contents
5.6. International Conference on Control, Automation and Systems
Contributed by: Hye-Soo Kim, conference@icros.org
2017 17th International Conference on Control, Automation and Systems (ICCAS 2017)
October 18(WED)-21(SAT), 2017
Ramada Plaza, Jeju Island, Korea
http://2017.iccas.org
2ND CALL FOR PAPERS:
http://icros.org/data/download/ICCAS2017/ICCAS2017 CFP.pdf
The aim of the ICCAS is to bring together researchers and engineers worldwide to present their latest works, and disseminate the state-of-the-art technologies related to control, automation, robotics, and systems.
Paper Submission: Authors are invited to submit regular papers (3 - 6 pages) or research poster papers (1 - 2 pages) to the website.
Indexed in: IEEE Xplore, EI compendex, SCOPUS
IMPORTANT DATES
Proposal for Invited/Organized Session (Mini-symposium)
- June 10, 2017: Submission deadline
Regular Papers (3 - 6 pages) & Invited/Organized Session Papers (1 - 6 pages)
- June 15, 2017: Submission deadline
- August 1, 2017: Notification of acceptance
- August 31, 2017: Submission of final camera-ready papers
Research Poster Papers (1 - 2 pages)
- August 22, 2017: Submission deadline
- August 31, 2017: Notification of acceptance
- September 7, 2017: Submission of final camera-ready papers
PLENARY SPEAKERS
- Richard D. Braatz (Massachusetts Inst. of Tech., USA)
- Reza Moheimani (Univ. of Texas, USA)
- Antonella Ferrara (Univ. of Pavia, Italy)
- Huijun Gao (Harbin Inst. of Tech., China)
- Atsuo Takanishi (Waseda Univ., Japan)
Organized by Institute of Control, Robotics and Systems (ICROS)
General Chair: Dong-il “Dan” Cho (Seoul Nat’l Univ., Korea / ICROS President) Organizing Chair: Doyoung Jeon (Sogang Univ., Korea)
Program Chair: Hyosung Ahn (GIST, Korea)
Back to the contents
5.7. Workshop on Networks and Control at University of Cambridge
Contributed by: Keith Glover, kg@eng.cam.ac.uk
WORKSHOP ON NETWORKS AND CONTROL, Wednesday 5 July 2017
Gonville and Caius College, University of Cambridge.
A workshop to mark the contributions of Malcolm Smith to the Control field on the occasion of his 60th birthday.
Invited speakers (confirmed):
M.Z.Q. Chen, P. Dewilde, T. Georgiou, S. Hara, T.H. Hughes, J.Z. Jiang, A. Rantzer, M. Vidyasagar, G.
Vinnicombe, Y. Yamamoto.
Chair: K. Glover
Website: http://www-control.eng.cam.ac.uk/Main/Workshop9
REGISTRATION DEADLINE: June 12, 2017
Back to the contents
5.8. IFAC World Congress Workshop: ”Rigidity Theory for Multi-agent Systems Meets Par¬allel Robots: Towards the Discovery of Common Models and Methods”
Contributed by: Daniel Zelazo, dzelazo@technion.ac.il
Call for participation
IFAC World Congress 2017 Workshop ”Rigidity Theory for Multi-agent Systems Meets Parallel Robots:
Towards the Discovery of Common Models and Methods”
Sunday, July 9, 2017 in Toulouse, France (full day)
https://parrigidwrkshp.sciencesconf.org/
https://www.ifac2017.org/
Important dates
15 April 2017: Early registration rates expire
9 July 2017: Workshop (full-day)
Overview & Topics
This workshop aims to explore connections and encourage discussion between two historically distinct com¬munities: the parallel robotics community and the multi-robot community. Although these two areas may appear as quite distant, they share a strong common underline theme: understanding how pairwise geo¬metrical constraints (e.g., relative distances or angles) can affect the mobility or state (pose) estimation for robotic systems. Moreover, there is a strong analogy between multi-agent systems and parallel robots: each robot of the system can be seen as a passive joint of a virtual mechanical (parallel) architecture and each measurement between two robots as a rigid connection between them, rigid connection whose dimension can vary thanks to a virtual actuator. So it is possible to find virtual parallel robot architectures associated with multi-agent systems.
Three types of sessions will be organized in order to promote interactivity / exchanges in the audience
- keynote sessions (6 invited speakers for a 35 minutes talk for each of them)
- interactive session
- panel discussion
We invite students and researchers to submit 1-page abstracts to be presented at an interactive session during the workshop. We encourage submissions related to the fields of multi-robot formation control and parallel robotics. Information can be found at the workshop website.
Invited Speakers
Hyo-Sung Ahn (A physical interpretation of the rigidity matrix), GIST Korea
Shiyu Zhao (Bearing-Based distributed control and estimation over robotic networks), University of Sheffield,
UK
Daniel Zelazo (Rigidity theory and formation control: A Tutorial), Technion, Israel
Jean-Pierre Merlet (Structural topology, singularity, and kinematic analysis), INRIA-Sophia, France
Marco Carricato (Screw theory and its application to robotics), Univeristy of Bologna, Italy
S´ebastien Briot (How theory on parallel robot singularities was used in order to solve sensor-based control
problems), LS2N-CNRS, France
Here are the schedule and keynote session abstracts: https://parrigidwrkshp.sciencesconf.org/
Organizers
Antonio Franchi, LAAS-CNRS, France
Daniel Zelazo, Technion, Israel
S´ebastien Briot, LS2N-CNRS, France
Paolo Robuffo Giordano, IRISA-CNRS, France
Registration at http://ifac2017.gipco-adns.com/
For more information and for submissions please contact parrigidwrkshp@sciencesconf.org
Back to the contents
6. Positions
6.1. PhD: University of Suttgart, Germany
Contributed by: Frank Allgower, allgower@ist.uni-stuttgart.de
PhD: University of Suttgart, Germany
Multiple PhD positions in Intelligent Systems, including Control, at the new International Max Planck Research School for Intelligent Systems in Stuttgart and Tubingen, Germany
The Max Planck Institute for Intelligent Systems and the Universities of Stuttgart and Tubingen are col-laborating to offer a new interdisciplinary Ph.D. program, the International Max Planck Research School for Intelligent Systems. This new doctoral program is starting in fall 2017 and will enroll about 100 Ph.D. students over the next six years.
This school is a key element of the state’s “Cyber Valley” initiative to accelerate basic research and com-mercial development in the broad field of artificial intelligence. Students are sought who want to earn a doctorate in the broad area of intelligent systems, including control systems.
The participating faculty are Frank Allgower, Matthias Bethge, Michael J. Black, Andres Bruhn, Peer Fis¬cher, Andreas Geiger, Philipp Hennig, Katherine J. Kuchenbecker, Hendrik Lensch, Georg Martius, Ludovic Righetti, Stefan Schaal, Bernhard Scholkopf, Metin Sitti, Alexander Sprowitz, Ingo Steinwart, Marc Tous¬saint, Ulrike von Luxburg, and Felix Wichmann.
Intelligent systems that can successfully perceive, act, and learn in complex environments hold great potential for aiding society. To advance human knowledge in this domain, we need doctoral students who are curious, creative, and passionate about research to join our school. Learn more at http://imprs.is.mpg.de
All aspects of the program are in English. You may join our program in late summer or early fall 2017. You will be mentored by our internationally renowned faculty. You will register as a university graduate student and conduct research for approximately three years. You can take part in a wide variety of scientific seminars, advanced training workshops, and social activities. Your doctoral degree will be conferred when you successfully complete your Ph.D. project. Our dedicated coordinator will assist you throughout your time as a doctoral student.
People with a strong academic background and a master’s degree in Engineering, Computer Science, Cogni¬tive Science, Mathematics, Control Theory, Neuroscience, Materials Science, Physics, or related fields should apply.
We seek to increase the number of women in areas where they are underrepresented, so we explicitly encourage women to apply. We are committed to employing more handicapped individuals and especially encourage them to apply. We are an equal opportunity employer and value diversity at our institutions.
Admission will be competitive. If selected, you will receive funding via an employment contract, subject to the rules of the Max Planck Society and the two participating universities.
In case of interest, please specify the desired main academic advisor with your application.
You can apply at http://imprs.is.mpg.de before midnight PST on April 17, 2017.
Back to the contents
6.2. PhD: Universit´e Laval, Canada
Contributed by: Andre Desbiens, desbiens@gel.ulaval.ca
PhD: Universit´e Laval, Canada
Three PhD positions are available at the LOOP (Laboratoire d’observation et d’optimisation des pro c´ed´es – Process Observation and Optimization Laboratory), Universit´e Laval, Qu´ebec City, Canada. The projects are in collaboration with the multinational biopharmaceutical Pfizer. They address industrial problems and the solutions will have significant impacts for Pfizer.
For pharmaceutical industries, automation and continuous processing is a way to become more competitive, to reduce production time, energy consumption and the amount of waste produced. Towards this objective, the projects are:
Project #1 - Coating of the tablets: development of an in-line vision sensor providing film-coating properties
(coating level, distribution across tablets, esthetical defects, etc.).
- Fractional factorial design
- Multivariate Image Analysis
- Partial Least Squares regression
- Validation of the machine vision sensor
Project #2 – Novel continuous drying of the granules (before they are compressed into tablets): safe and
robust in-line minimization of the drying time and/or energy consumption while insuring a desired final
humidity of the particles and avoiding their overheating.
- First-principles modelling and model calibration
- State estimation
- Model predictive control
- Real-time optimization
Project #3 - Freeze-drying of vials: safe and robust in-line minimization of the primary drying time and/or energy consumption while insuring that sublimation is completed and avoiding to exceed the collapse tem
perature.
- First-principles modelling and model calibration
- State estimation
- Model predictive control
- Real-time optimization
- Heating policies for various vials arrangements
The final stage of the three projects is to implement and validate the most promising approaches on pilot units.
Candidate profile:
- should have completed, or about to complete, a MSc degree in Electrical or Chemical Eng., or related
areas,
- strong background in multivariate statistics and/or first-principles modelling and/or systems and control,
- solid programming skills in Matlab,
- ability to work in multi-disciplinary teams,
- excellent communication skills (oral and written) in English - a plus if knowledge of French (courses are
given in French).
Please send a complete CV, a motivation letter and transcripts to Prof. Andr´e Desbiens (desbiens@gel.ulaval.ca)
with the subject ”E-Letter PhD position”.
Back to the contents
6.3. PhD: University of Agder, Norway
Contributed by: Jing Zhou, jing.zhou@uia.no
PhD Position at University of Agder, Norway
Supervisor: Prof. Jing Zhou
Project Topic: Coupled Dynamics Between Vessel and Crane
The University of Agder invites applications for a PhD fellowship in Coupled Dynamics Between Vessel and
Crane. The position is linked to the Department of Engineering Sciences and the contract is for a period of
3 years. Starting date in September 2017 or negotiated with the faculty. The position will also be linked to
work-package 3 in the research centre SFI Offshore Mechatronics.
Deadline for application: 31 May 2017.
Brief Description of the Research Project
With the increased focus on offshore operations to deep water fields, it is important to perform crane operations faster with increased weather operation window. In addition, necessary precautions towards the safety of human lives, environment and property should be taken. To achieve satisfactory control performance, the investigation of coupled dynamics between vessel and crane is very important. The accurate control for the dynamic positioning of the payload in a vessel-mounted crane system is challenging due to the exogenous disturbances such as actuator movement, accidental collisions, the motion of the vessel induced by waves and ocean currents, and so on. The objective is to increase the control performance of the crane and vessel, and ensure the safety of operations.
The applicant should have:
• Strong academic credentials, written and spoken English proficiency
• Strong knowledge in systems and control, mechatronics, mathematics or marine systems
• Good academic record, showing a strong theoretical/mathematical background and a strong interest in
dynamical control systems
• Strong programming skills and years of experience in using numerical tools such as MATLAB
Remuneration
The position is remunerated according to the State salary scale, code 1017 Research Fellow, salary NOK 435 500 gross per year. A 2 % compulsory pension contribution to the Norwegian Public Service Pension Fund is deducted from the pay according to current statutory provisions.
For further information please contact Professor Jing Zhou, tel. +47 37 23 3191, e-mail: jing.zhou@uia.no.
Detailed information and application procedure:
https://www.jobbnorge.no/ledige-stillinger/stilling/137288/phd-research-fellow-in-instrumentation-and-real-time-control-of-long-reach-light-weight-arm-intended-for-use-offshore
Back to the contents
6.4. PostDoc: University of California at San Diego, USA
Contributed by: Miroslav Krstic, krstic@ucsd.edu
Postdoctoral position in control of ELECTRIC MOTOR DRIVES at UNIVERSITY OF CALIFORNIA, SAN DIEGO
A postdoctoral position is open, with the intent of it being filled in/by Fall 2017, at University of Califor¬nia, San Diego, under Professor Miroslav Krstic, on the topic of control of large electric motor drives in collaboration with General Atomics, a major local San Diego company. The research challenges include the achievement of stable operation in the face of significant variations in the sampling rates and delays. The anticipated industrial impact is in oil/gas industry.
The appointment is for one year, with the possibility of extension subject to availability of funds and strong performance. The salary is in accordance with the University of California postdoctoral salary scale, which is anticipated to be in the high-$40K’s range for the 2016-2017 academic year.
Eligibility and requirements. Only candidates who are PERMANENT RESIDENTS OF THE UNITED STAES can be appointed for this position. The required background, in addition to strong training in control systems, includes modeling and CONTROL OF ELECTRIC MACHINES and power electronics.
UC San Diego’s Cymer Center for Control Systems and Dynamics houses one of the world’s premier research groups and training programs in control engineering.
Interested candidates should contact Professor Krstic (krstic@ucsd.edu; http://flyingv.ucsd.edu) and include their detailed CV, information on their current and near-term job status, US immigration/residency sta¬tus, and a list of references. No response will be sent to applications that do not meet the capitalized REQUIREMENTS mentioned above.
Back to the contents
6.5. PostDoc: CNRS – CentraleSup´elec – Univ. Paris-Sud – Univ. Paris-Saclay, France Contributed by: Antoine Girard, antoine.girard@l2s.centralesupelec.fr
Postdoctoral Positions - Towards programmable cyber-physical systems: a symbolic control approach
Supervisors: Antoine Girard (Antoine.Girard@l2s.centralesupelec.fr)
Location: Laboratoire des Signaux et Syst`emes - L2S, CNRS – CentraleSup´elec – Univ. Paris-Sud – Univ.
Paris-Saclay, Gif-sur-Yvette, France
Duration: One year, starting September 2017, with possibility for one additional year
Two postdoctoral positions are opened within the PROCSYS project, funded by an ERC Consolidator Grant.
Context and Objectives:
Cyber-physical systems (CPS) consist of computational elements monitoring and controlling physical en¬tities. The main objective of the PROCSYS project is to propose a general framework for the design of programmable CPS that will allow engineers to develop advanced functionalities using a high-level language for specifying the behaviours of a CPS while abstracting details of the dynamics. Controllers enforcing the specified behaviours will be generated from a high-level program using an automated model-based synthesis tool. Correctness of the controllers will be guaranteed by following the correct by construction synthesis paradigm through the use of symbolic control techniques: the continuous physical dynamics is abstracted by a symbolic model, which is a discrete dynamical system; an interface consisting of low-level controllers is designed such that the physical system and the symbolic model behaves identically; a high-level symbolic controller is then synthesized automatically from the high-level program and the symbolic model.
Work description:
We will develop a high-level language, based on the formalism of hybrid automata, which will enable to specify rich behaviours while enabling the development of efficient controller synthesis algorithms. The project will also tackle two bottlenecks in the area of symbolic control. Firstly, scalability issues will be addressed by the computation of more compact symbolic models and by controller synthesis algorithms that require only partial exploration of the symbolic models. Secondly, robustness will be ensured at all levels of control by developing novel algorithms for the synthesis of robust interfaces and of symbolic controllers.
Background of the candidate:
The candidate must hold a PhD in control theory or computer science with a strong mathematical back¬ground. A prior experience in the area of hybrid systems is recommended. Programming skills are also needed. Applications must include a cover letter, a detailed CV, the preprints of the two most significant publications, and two references who may be asked to provide letters of recommendation.
All documents should be sent in a single pdf file to the following email addresses: Antoine.Girard@l2s.centralesupelec.fr
Back to the contents
6.6. PostDoc: Universidad T´ecnica Federico Santa Mar´ıa, Chile
Contributed by: Juan I. Yuz, juan.yuz@usm.cl
POSTDOCTORAL POSITION at AC3E, UTFSM, CHILE
The Advanced Center for Electrical and Electronic Engineering (AC3E) was created on 2014 to group individual research efforts into multi- and inter-disciplinary teams and re-focus research towards industry related problems to spark innovation. The center is hosted by Universidad T´ecnica Federico Santa Mar´ıa (UTFSM), in Valpara´ıso, Chile, one of the top engineering universities of Chile. Additional information about the center can be found at www.ac3e.usm.cl and the university can be found at www.usm.cl
We are looking to hire a highly qualified individual as postdoctoral researcher for the area of Intelligent Transportation systems. In particular, we want to focus on the development of Control and Communications techniques applied to the deployment of platoons and their interaction with different environments. As a consequence, it is desirable that the applicants have a demonstrable background in Control Systems, System Identification and Wireless Communications. Nevertheless, we are also interested in candidates with expertise in other research areas in Electrical and Electronic Engineering, including:
• System Identification,
• Nonlinear Systems modeling and control,
• Fault Diagnosis and Prognosis,
• Networked and Distributed Control Systems,
• Performance Limitations and Control Design,
• Control of Partial Differential Equations.
The researchers associated to the Control and Automation line are:
• Dr. Juan C. Ag¨uero, UTFSM (Line of Research Leader),
• Dr. Juan Yuz, UTFSM,
• Dr. Marcos Orchard, U. de Chile,
• Dr. Alejandro Rojas, U. de Concepci´on,
• Dr. Eduardo Cerpa, UTFSM,
• Dr. Francisco Vargas, UTFSM.
Required Documents
1. Curriculum Vitae.
2. List of academic productivity (publications, books, patents).
3. Cover letter stating your interests and why you want to become part of AC3E.
4. Documentation accrediting the possession of a PhD or Doctoral degree.
5. Letter of reference or a list of 2 referees that might be contacted.
*Please provide all documents in one PDF file.
General Information
• The duration of the Postdoctoral fellow will be up to 2 years.
• Applications and inquiries should be sent to the following email: ac3e@usm.cl
Back to the contents
6.7. PostDoc: Sandia National Laboratories, USA
Contributed by: Chimene Lopez, recruitingads@sandia.gov
Sandia National Laboratories is seeking a POSTDOCTORAL APPOINTEE for our Albuquerque, NM site.
Please review job description below. Apply by completing an online application at:
http://www.sandia.gov/careers
Click on “View all jobs” and enter Job Numb er“657239” into the Keyword search.
On any given day, you may be called on to:
Postdoctoral researcher (or appointee) in dynamical systems analysis and control systems design.This po¬sition requires an active researcher in basic and applied research for marine hydrokinetic (MHK) energy systems. The selected candidate must have the ability to contribute to a diverse research portfolio and also support execution of the program for our primary sponsor. The selected candidate will have the opportunity to actively publish and present his/her research at conferences and workshops and in technical journals. The selected candidate must also be able to interface with the other activities in the department to assist in furthering the development of a cross-cutting research portfolio.
Required:
• Ph.D. in Electronic, Mechanical, Aerospace Engineering or similar engineering
• Experience with dynamic system analysis, state-estimation, and control system design
• Experience in signal processing and data analysis.
Desired:
• Test equipment development experience
• Experience in design and implementation of real time control systems
• Experience in specification, design, construction/implementation of sensor systems for feedback control
• Experience in system identification
Department Description:
Sandia’s Water Power Technologies Department performs research and development to improve performance, lower costs, and accelerate the deployment of water power technologies. The primary focus area of the department’s work is Marine Hydrokinetics (MHK) research with lower levels of effort in Offshore Wind (OW) and Hydropower. The department is responsible for all aspects of Marine Hydrokinetic device design, performance, system reliability, system integration, and environmental evaluations for the Department of Energy¿s Wind and Water Power Technologies Office and in collaboration with industry partners. Offshore Wind Energy activities currently include Vertical Axis Wind Turbine (VAWT) design and analysis and foundation scour. Hydropower efforts focus on seasonal simulation and optimization. Through partnerships with industry, academia, other National Labs, and public dissemination of results, Sandia serves an important role in energy security for the nation
About Sandia:
Sandia National Laboratories is the nation’s premier science and engineering lab for national security and technology innovation, with teams of specialists focused on cutting-edge work in a broad array of areas. Some of the main reasons we love our jobs:
Challenging work with amazing impact that contributes to security, peace, and freedom worldwide
Extraordinary co-workers
Some of the best tools, equipment, and research facilities in the world
Career advancement and enrichment opportunities
Flexible schedules, generous vacations,strongmedical and other benefits, competitive 401k, learning oppor
tunities, relocation assistance and amenities aimed at creating a solid work/life balance*
World-changing technologies. Life-changing careers. Learn more about Sandia at: http://www.sandia.gov
*These benefits vary by job classification.
Security Clearance:
No clearance required.
This position does not currently require a Department of Energy (DOE)-granted security clearance.
Sandia will conduct a pre-employment background review that includes personal reference checks, law en-forcement record checks, and employment and education verifications. Further, employees in New Mexico must pass a U.S. Air Force background screen for access to the work site. Substance abuse or illegal drug use, falsification of information, criminal activity, serious misconduct or other indicators of untrustworthiness can cause access to be denied or terminated, rendering the inability to perform the duties assigned and resulting in termination of employment.
If hired without a clearance, and one subsequently becomes required or you bid on positions that require a DOE-granted security clearance, a pre-processing background review that includes personal reference checks, law enforcement record and credit checks, and employment and education verifications may be conducted prior to a required federal background investigation. Applicants for DOE-granted security clearances must be U.S. citizens and be able to obtain and maintain the appropriate DOE security clearance as required for the position.
EEO Statement:
Equal opportunity employer/Disability/Vet/GLBT
Back to the contents
6.8. PostDoc: Israel Institute of Technology
Contributed by: Tal Shima, tal.shima@technion.ac.il
A post-doctoral position is available at the Department of Aerospace Engineering, Technion - Israel Institute of Technology, in Haifa, Israel.
The research is in the general area of guidance of unmanned vehicles, mainly aerial ones. The scope of the research is broad and possible topics include: 1) Pursuit-evasion guidance; 2) Cooperative guidance; 3) Intertwined guidance and flight control; 4) Intertwined guidance and estimation; 5) Intertwined guidance and task assignment. The research will involve both theoretical and algorithmic aspects. Laboratory experiments on available ground and aerial robots may also be performed.
Candidates for this position should have a Ph.D. in engineering (aerospace, mechanical, electrical, or similar), computer science, or applied math. Strong background in optimal control and/or differential games is an advantage.
Application material should include:
- a cover letter
- detailed curriculum vitae, including educational background and a list of publications
- undergraduate and graduate studies grades transcripts
- contact information for at least three academic references
The material should be submitted, in pdf, via e-mail, to Prof. Tal Shima, tal.shima@technion.ac.il The position is available immediately and applications will be handled as they arrive until the position is filled.
For further inquiries, please contact Tal Shima at: tal.shima@technion.ac.il
Back to the contents
6.9. PostDoc: University of Florida, USA
Contributed by: Michael McCourt, mccourt@ufl.edu
PostDoc: University of Florida, USA
Contributed by: Michael McCourt, mccourt@ufl.edu
Postdoctoral Research Fellowship in Control and Robotics
Location: UF REEF, Shalimar, FL
Department: Mechanical and Aerospace Engineering
Salary rate: $65,000 annual, 1.0 FTE
Position Description:
The University of Florida REEF facility is announcing a post-doctoral fellowship sponsored by the Air Force Research Laboratory, Munitions Directorate (AFRL/RW) at Eglin AFB, Florida in the area of control and estimation. This AFRL/RW sponsored project, “Privileged Sensing Framework”, has focused on coopera¬tive control and estimation in human-machine systems. This project has investigated novel approaches in fusing human perceptions with autonomous sensor measurements to reduce estimation error and improve coordination in human-machine teams. Desired skills for this postdoc position include prior experience with Bayesian estimation, classification, novel representations of knowledge, risk-based decision making, control systems, learning in autonomous control, and distributed control architectures. While this position is fo¬cused primarily in the development of original theory, some applied demonstrations are encouraged in either simulation or hardware utilizing ground robots and quad rotors in a motion capture lab space.
The UF REEF facility is located in Shalimar, FL. It is a collaborative space with researchers from UF, AFRL, and other institutions. Research projects are in a variety of fields including control and estimation,
communication, coordination in autonomous vehicles, industrial optimization, and materials science. Col¬laborative work between these groups is highly encouraged. There are opportunities for robotic experiments utilizing ground robots and quad rotors in a motion capture lab space.
Qualifications:
Candidates must hold a Ph.D. in a field closely related to control and estimation and a track record of conducting and publishing quality research at top conferences and in scientific journals. This position requires good communication and collaborative research skills as this research group is made up of both UF and AFRL researchers.
Information:
For more information or to apply, please contact Dr. Michael McCourt, mccourt@ufl.edu
Back to the contents
6.10. PostDoc: National Institute of Informatics, Japan
Contributed by: Ichiro Hasuo, i.hasuo@acm.org
PostDoc: National Institute of Informatics, Japan
For our new 5-year research project (ERATO MMSD, Metamathematics for Systems Design) we are looking for senior researchers and postdocs (10+ positions in total and several are still open), together with research assistants (PhD students) and internship students.
This broad project aims to extend the realm of formal methods from software to cyber-physical systems (CPS), with particular emphases both on logical/categorical metatheories and industrial application esp. in automotive industry. The project covers diverse areas that include: control theory, control engineering, formal methods, programming languages, software science, software engineering, machine learning, numerical optimization, user interface, mathematical logic and category theory.
For more about the project please visit http://group-mmm.org/eratommsd
About the open positions
http://group-mmm.org/eratommsd/openpositions.html
has more information (including how to apply/inquire).
Back to the contents
6.11. PostDoc: Queen Mary University of London, UK
Contributed by: Guang Li, g.li@qmul.ac.uk
The school of Engineering and Materials Science at Queen Mary University of London has a vacancy for a postdoctoral research assistant to work on an EPSRC funded project “Control of Launch and Recovery in Enhanced Sea-States”. This project will be carried out in the Division of Engineering Science.
The practical driver of this project is to extend the range of sea-states in which existing wave limited maritime operations can be safely carried out. Important examples of these operations are launch and recovery (L&R) from a mother-ship of small craft, manned and unmanned air vehicles and submersibles. In this project the research aim is to develop a novel approach to predicting a suitable time instant at which to initiate a L&R operation, together with a confidence measure (provided as advice to a human operator), and then to control the execution of the subsequent lift operation once initiated using a form of Model Predictive Control (MPC). The successful applicant will possess a relevant PhD or equivalent qualification/experience in a related field of study and will have recognised expertise in the areas of system modelling and control and possess in-depth understanding of this specialism to enable the development of new knowledge and
understanding within the field. Applicants will possess proven expertise in the areas of systems modelling and control and highly developed skills in Matlab & Simulink. The successful applicant will 1) develop fast MPC for the L&R process, 2) conduct stability analysis with preview information and varying constraints, 3) conduct co-design to integrate system parameter selection into controller design, 4) present results in project meetings, workshops and conferences, 5) publish high-quality papers.
The post is a full-time, fixed term appointment for 36 months and is available from 1 July 2017. Starting salary will be £36,064 per annum, inclusive of London Allowance. Benefits include 30 days annual leave, defined benefit pension scheme and interest-free season ticket loan.
Informal enquiries should be addressed to Dr Guang Li at g.li@qmul.ac.uk or on +44 (0) 020 7882 6116.
Details about the school can be found at www.sems.qmul.ac.uk
Back to the contents
6.12. PostDoc: CNRS, France
Contributed by: Hannah Walter, Hannah-Christina.Walter@gipsa-lab.fr
Scale-Free Modelling and Control for Large Scale Complex Networks
Supervisors: Carlos Canudas-de-Wit (DR-CNRS, supervisor), Sandro Zampieri (co-supervisor).
Context: ERC-AdG Scale-FreeBack (see http://scale-freeback.eu)
Application type: Post-do c. Gross salary: 2530-3509(if more than 2 year after PhD) Euros/M
Duration: 12+12 months. Employer: CNRS. Location: Grenoble, France
Applications: http://scale-freeback.eu/openings/
Context. Scale-FreeBack is an ERC Advanced Grant 2015 awarded to Carlos Canudas-de-Wit, Director of Research at the National Center for Scientific Research, (CNRS), during Sept. 2016-2021. The ERC is hosted by the CNRS. The project will be conducted within the NeCS group (which is a joint CNRS (GIPSA-lab)-INRIA team). Scale-FreeBack is a project with ambitious and innovative theoretical goals, which were adopted in view of the new opportunities presented by the latest large-scale sensing technologies. The overall aim is to develop holistic scale-free control methods of controlling complex network systems in the widest sense, and to set the foundations for a new control theory dealing with complex physical networks with an arbitrary size. Scale-FreeBack envisions devising a complete, coherent design approach ensuring the scalability of the whole chain (modelling, observation, and control). It is also expected to find specific breakthrough solutions to the problems involved in managing and monitoring large-scale road traffic networks. Field tests and other realistic simulations to validate the theory will be performed using the equipment available at the Grenoble Traffic Lab center, and a microscopic traffic simulator replicating the full complexity of the Grenoble urban network. The proposed work will be undertaken in the context of this project.
Topic description. This research proposal deals with the problem of setting up a suitable modelling framework for complex systems corresponding to large-scale networks. The original system is assumed to describe a homogenous network in which the node/link distribution is a bell-shaped, exponentially decaying curve. Homogenous networks cover many critical systems of interest (such as road traffic networks, power grids, water distribution systems, etc.), but are inherently complex. Scale-FreeBack is elaborated on the idea that complexity can be broken down by abstracting an aggregated scale-free model (represented by a network with a power law degree distribution), by merging/lumping neighboring nodes in the original network. In that, supper-nodes (nodes with a lot of connections) are created and represented by “aggregated” variables. Controlling only boundary inputs and observing only aggregated variables allows to cut-off the system complexity. The following questions will be addressed:
1) Defining the most suitable level of aggregation for the model. This boils down to defining and sizing the state-vector, the control inputs and outputs. A first question is how to define the right level of aggregation, and investigate new metrics trading quantifiers reflecting an optimal level of scalability (a suited node/link distribution) of the associated network graph, with other performance indexes reflecting the system’s closed-loop operation.
2) The second question focuses on how the aggregation process, in addition to the scale-free property, will yield models consistent with the design of control and the observation goals. The aggregation process will have to include observability and controllability properties which are consistent with the evolutionary nature of scale-free aggregated models (aggregation process is evolutionary in the sense that the network changes and so the aggregated modules will change accordingly while preserving the scale-free properties).
3) Finally, innovative concepts such as peripheral controllability (i.e. controlling the boundary flows in a lumped node rather than controlling each single node separately), and energy-weighted controllability metrics (where controllability is qualified by assessing the energy costs as a function of the controllable nodes [Zam-et-al’14]) will be extended in this project to the context of scale-free models. While only open loop metrics have been considered so far, we aim to propose new closed loop metrics also taking inspiration from road traffic networks application. Finally, we will propose and investigate different new weak notions of controllability in which the controllability is determined with respect to a limited subspace (peripheral and/or sparse controllability), and to devise the associated control methods.
Request Background. Control Systems, Applied mathematics.
Applications. Please follow the application procedure indicated at (http://scale-freeback.eu/openings/)
Back to the contents
6.13. PostDoc: UT-Dallas, USA
Contributed by: Reza Moheimani, reza.moheimani@utdallas.edu
We are seeking a postdoctoral research fellow to join our multidisciplinary research group, based in the Laboratory for Dynamics and Control of Nanosystems at UT-Dallas. The applicants are expected to have a PhD in a relevant field (or be close to completion), have a strong analytical background and be familiar with advanced control system design techniques. Familiarity with analog electronics design and rapid prototyping systems is a major plus. Familiarity with scanning probe techniques is highly valued.
To be considered, the applicants should send their CVs including a list of publications and names and
addresses of three referees to D. Reza Moheimani (contact email: Reza.Moheimani@utdallas.edu ).
Back to the contents
6.14. Research Fellow/Associate: National University of Singapore, Singapore Contributed by: Chong Jin Ong, mpeongcj@nus.edu.sg
Applicants are invited for Research Fellow/Associate positions to work on approaches to distributed con-trol/optimization of multi-agent system. In particular, effective approaches are sought that solve the con¬sensus problem for a multi-agent, network system under several settings: the presence of global constraint, time-switching network and/or state and control constraints. The position is with the Department of Me¬chanical Engineering, National University of Singapore. Applicants should possess at least a Master’s degree with at least 2 years’ relevant work experience for Research Associate position. Candidate with PhD de¬gree (preferably in Multi-agent Control, distributed optimization) will be considered for Research Fellow position. The applicants should have very good foundation in mathematics and control theory. Those who have recently obtained a PhD degree in general control theory, computer science and mathematics are also
encouraged to apply. Current PhD students who are on the last leg of their candidature (submitting their thesis within the next few months) or waiting for their oral defense may also be considered.
Remuneration will commensurate with experience.
Applications with full CV to be submitted electronically to Assoc Prof Ong Chong Jin at the email:
mpeongcj@nus.edu.sg
Back to the contents
6.15. Faculty: Sharif University of Technology, Iran Contributed by: Amin Nobakhti, nobakhti@sharif.edu
Sharif University of Technology is pleased to announce openings for tenure track positions and is inviting highly qualified applicants to join the department of electrical engineering in areas related to control system engineering.
About the university
Sharif University of Technology is Iran’s premier technical university with a distinguished track record in scientific research and discovery. The department of Electrical and Electronics is especially renowned nationally and inter-nationally for its outstanding faculty and the extremely talented students. For more information on the university and the department please visit http://www.sharif.ir/web/en.
Who should apply?
We are seeking graduates with a Ph.D. in control systems or related files from a top ranking international university and with a portfolio of fundamental and seminal research contributions to the field. A postdoc is not necessary, but it is clearly preferred. A good experience in student supervision and teaching, together with experience of applied and industrial research are essential. Successful candidates are expected to develop innovative courses in the area of control systems.
Research areas
Applicants are welcome in all areas related to control system, but are especially sought after in the following
areas:
-Big Data and machine Learning
-Complex networks
-Systems biology
-Economic systems
-Nonlinear systems
-Robotics
-Discrete-event systems
-Hybrid systems
-Renewable energy systems
What is required in an application pack?
-A covering letter
-A full academic CV
-Statement of Teaching
-Statement of Research
-Personal Statement
-Transcript of grades for B.Sc. , M.Sc., and Ph.D. degrees
-Details of three references
Interested candidates should send their application pack for consideration to nobakhti@sharif.edu. All in
quiries should be addressed to nobakhti@sharif.edu
Back to the contents
6.16. Faculty: University of Louisiana at Lafayette, USA
Contributed by: Afef Fekih, afef.fekih@louisiana.edu
Assistant Professor, Electrical & Computer Engineering (Tenure Track) University of Louisiana at Lafayette, USA
We are seeking candidates with a robust academic record, who will be active educators in their fields. Tenure-track faculty are expected to be excellent teachers and to develop strong funded research programs. In addition, serve as a mentor to students, work with colleagues to assess and improve curricula and demonstrate institutional citizenship through active engagement at the department, college, university, community, and professional levels.
QUALIFICATIONS:
A Ph.D. in Electrical & Computer Engineering or closely related field, preferably with a B.S.E.E. from an ABET accredited program.
Successful candidates must be committed to working effectively with diverse student populations. Applicants are expected to describe their commitment to fostering a diverse educational environment through their research, teaching, and/or service activities.
Please follow the link below for further details:
https://www.higheredjobs.com/institution/details.cfm?JobCode=176446670&Title=Assistant%20Professor%20%28Tenure%
Back to the contents
6.17. Faculty: Zhejiang University of Technology, China
Contributed by: Qiu Xiang, qiuxiang@zjut.edu.cn
Faculty Position: Zhejiang University of Technology , Hangzhou, China http://www.auto.zjut.edu.cn/WebSite/Job/JobList.aspx
Zhejiang Control Science and Engineering First-Class (Class A) Discipline Recruitment Announcement
Zhejiang University of Technology (ZJUT), sitting by the beautiful West Lake, Hangzhou, is a Zhejiang Province and the Ministry of Education co-supported, provincially governed key university, who owns one of the only 14 Collaborative Creation Centers in the first initiative of the state “2011 Program”. ZJUT has its beautiful campus covering more than 3000 mu, which accommodates 24 Colleges, more than 37,000 full-time students and more than 3,300 staffs. ZJUT is proudly to have 2 self-owned and 2 sharing Fellows of the Chinese Academy of Engineering, as well as more than 1400 faculties with senior professional titles. ZJUT has State Key Disciplines, State Engineering Research Centers, State University Science Parks, Centers for Postdocs, as well as the power of awarding Doctors, Masters, MBAs and recruiting foreign students and those from Hong Kong, Macao and Taiwan.
The Control Science and Engineering Discipline within the College of Information Engineering was one of the Priority-among-Priorities Disciplines (selected by Zhejiang Provincial Government in 2009), and is now one of the Zhejiang First-Class (Class A) Disciplines in the first initiative of the Program in 2015. The Discipline now has the Doctoral Program at the first-level discipline, the Center for Postdocs, and the Zhejiang Collab¬orated Key Laboratory of Embedded Systems. The College of Information Engineering where the Discipline
is in has 5 undergraduate programs: Automation, Electrical Engineering and Its Automation, Electronic In-formation Engineering, Communication Engineering, and Electronic Science and Technology. The Discipline is now recruiting faculties in the following areas at the levels of State and Zhejiang Provincial “1000 Plan” high-level talents, Zhejiang “Qianjiang Scholars”, ZJUT “Yunhe Specially-Appointed Professors”, “ZJUT Professors”,outstanding PhDs and postdocs, etc.
(1) Control Science and Engineering, including advanced control theory, robotics, machine vision, pattern recognition, industrial networked control systems, MES, etc.
(2) Electrical Engineering, including electric drive, power electronics, new energy, etc.
(3) Mechatronic Engineering, including high-precision servo control of mechatronic devices, the modelling and dynamic analysis of robots, etc.
(4) Computer Science and Technology, including smart city, smart healthcare, big data, cloud computing, IoT, industrial control software, etc.
A. Selection criteria
High-level talents (Chang jiang Scholars, 1000 Plan Scholars, Qianjiang Scholars, etc.) You have major achievements and influence in your research area that have already been recognized by national and interna-tional researchers, or have great potentials of future development; You also meet the criteria of corresponding talents programs. ZJUT Professors /Associate Professors You have a PhD degree obtained from a recog-nized university or research institutes with at least one year of oversea research experience in a well-known foreign institute; You have research achievements recognized by national and international researchers; Your application also passes the review process at the university level (ZJUT).
Outstanding PhDs/Postdocs You have a PhD degree obtained from a recognized university or research in-stitute; You have high-quality research outputs and the professional skills required by a university lecturer, and great potentials of your future career.
B. Salary and welfare
(1)National-Level Top Tier Talents: Fellows of Chinese Academy of Sciences or Chinese Academy of En-gineering, “Special Support Program” Distinguished Talents, Principal Investigators of NSFC Innovative Research Team, or other talents at the equivalent level. Treatment:Negotiation on the case by case basis.
(2) National-Level Top Tier Talents:National “1000 Plan” Scholars (long-term), Changqiang Scholars, NSFC Distinguished Young Scholars, “Special Support Program” Outstanding Talents, winners (rank first) of three major national science awards, or other talents at the equivalent level. Salary (CNY):> 700K /Year; Housing Benefit(CNY):3M-5M; Startup Funds(CNY):Case by case.
(3) National-Level Young Talents:“Special Support Program” Outstanding Young Talents, “1000 Plan” Young Scholars, “Chang jiang Young Scholars, NSFC Outstanding Young Scholars, 973 Program Young Scholars, “Millions of Talents Program” Scholars, or other talents at the equivalent level. Salary (CNY):> 450K /Year; Housing Benefit(CNY):1.5M-2.5M; Startup Funds(CNY):1M-3M.
(4) Provincial-and-Ministry-Level Talents,Yunhe Specially-Appointed Professors:CAS “100 Plan” Scholars, Zhejiang ”Qianjiang Scholars”, Zhejiang “1000 Plan” (long-term) Scholars, or other talents who have made significant academic contributions with great potentials of development and who are awarded “Yunhe Specially-Appointed Professors” after the review of ZJUT. Salary (CNY):> 350K /Year; Housing Bene-fit(CNY):1.5M; Startup Funds(CNY):0.5M-1M.
(5) ZJUT Professors,ZJUT Associated Professors:You have a PhD degree obtained from a recognized uni-versity or research institutes with at least one year of oversea research experience in a well-known foreign institute; You have research achievements recognized by national and international colleges; Your applica¬tion also passes the review process at the university level. Salary (CNY):Salaries at the appropriate levels;
Housing Benefit(CNY):0.4M-0.5M; Startup Funds(CNY):0.1M-0.2M.
(6) Outstanding PhDs/Postdoctors: You have a PhD degree obtained from a recognized university or re-search institute; You have high-quality research outputs and the professional skills required by a university lecturer, and great potentials of your future career. Salary (CNY):Salaries at the appropriate levels; Housing Benefit(CNY):0.3M.
(7) Postdocs (leading to a faculty): Besides the basic salary and welfare, 50K/Year subsidy is provided for the first two years, with the possibility of continuing this subsidy plus a one-off 200K housing benefit if you are accepted to ZJUT public institution business unit.
C. Required documents
(1) One self-recommendation letter covering your study and professional records, your teaching and research statements, your achievements, your work plan as well as your possible requirements from us.
(2) A list of your research funds, awards, and publications in the recent five years.
D. Contact us
Dr. Qiu,
Email : qiuxiang@zjut.edu.cn
Mobile: +86-13867469319
Address: Xiaoheshan College Park, College of Information Engineering, Zhejiang University of Technology,
310023
Zhejiang Control Science and Engineering First-Class (Class A) Discipline
Apr 9, 2017
Back to the contents
6.18. Faculty: University of Newcastle, Australia
Contributed by: Bj¨orn R¨uffer, bjorn.ruffer@newcastle.edu.au
The University of Newcastle, Australia is soliciting applications for a new full professor in applied mathe¬matics. Mathematical systems theory and optimization are research areas of particular interest, but other areas of applied mathematics are equally welcome.
Application deadline: May 12, 2017
Location: Newcastle, Australia (about 170km north of Sydney)
University website: http://www.newcastle.edu.au
Position description:
https://hronline.newcastle.edu.au/pls/alesco/WK8127$VAC.QueryView?P VACANCY REF NO=3249
Back to the contents
H(x) = arg max(
E=1 αt[ht(x) = Ci])
Z=~ cD(i)exp(αI[h(x) = y])
qigre mlin ®ts t~M
and the rightside shaded represents samples mis-classified. On each side, samples are grouped by class labels, i.e., C1, C2 and C3. By weighting and normalizing of Ad-aBoost.M1, correctly classified space shrinks and misclas-sified space expands untilthese two parts getto equal. Fig-ure 2 (b) demonstrates this result. The notable point is on each part, correctly classified and misclassified, each group (class)shrinks orexpands atthe same ratio.Observa-tionally,the classes with relatively more misclassified sam¬ples willgetexpanded,which are notnecessary the classes we care about. To strengthen the learning on the “weak” classes,we expectmore weighted sample sizes on them.
5.2 AdaC2.M1
The learning objective of AdaC2.M1 algorithm is to se-lectαt for minimizing Zt on each round (Equation 30).The first derivative ofZt as a function ofαt is
dZ
Zt(α) =
dα
= ciDt(i)I[ht(xi) = yi]exp(αtI[ht(xi) = yi])
i
= Zt Dt+1(i)I [ht(xi) = yi]
i
by definition ofD(t+1) (Equation 26). To minimize Zt,αt is selected such thatZ'(α) = 0. The unique solution for αt is presented by Equation 31. By the definition ofDt+1 (Equation 26),for the nextiteration we willhave
E ~Dt+1(i) = Dt+1(i) (34)
i,h(x)=y i,h(x)=y
It indicates that weight of the correctly classified group and that of the misclassified group get to even after sam-ple weights being updated by AdaC2.M1,i.e.,TP = FP. Same as AdaBoost.M1,this weighting resultwillmake the learning of nextiteration the mostdifficult.
By the weight update formula of AdaC2.M1( Equation 26), sample weights of two groups respecting to class i, TP(i) and FP(i), updated from the tth iteration to the (t + 1)th iteration can be summarized asTPt+1(i) = c(i) •
qigre lin ®ts ®
T Pt(i)/eαt and FPt+1(i) = c(i) •FPt(i) • eαt. Wherec(i) denotes the misclassification costof classi.This weighting process can be interpreted in two steps. At the first step, each sample,no matter in which groups (TP or FP),is first weighted by its costitem (which equals to the misclassifica-tion cost of the class that the sample belongs to). Samples of the classes with larger costvalues willobtain more sam-ple weights, on the other side, samples of the classes with smaller cost values will lose their sizes. Consequently,the class with the largestcostvalue willalways enlarge its class size atthis phase.The second step is actually the weighting procedure of AdaBoost.M1, i.e., weights of false predic-tions are expanded and those oftrue predictions are shrunk. The expanding or shrinking ratio for samples of allclasses is the same.
To demonstrate this weighting process,we use the same example as illustrated for AdaBoost.M1. In this case, we associate each class with a misclassification cost. Suppose the costs are 3,1 and 2 respecting to class C 1,C 2 and C 3. Each sample obtains a cost value according to its class la-bel.Letthe sample distribution aftera classification process presented in Figure 3(a) be the same with thatpresented in Figure 2(a). By the weighting strategy of AdaC2.M1, the firststep is to reweight each sample by its costitem. After normalizing,classes with relative larger costvalues are ex-panded,oppositely,the other class is shrunk. In our exam-ple,class sizes of class C 1 and C 3 are increased and class size of class C 2 is decreased as presented in Figure 3 (b). Atthe nextstep,correctly classified space shrinks and mis-classified space expands until these two parts get to even. If we compare Figure 3(c) with Figure 2 (b),obviously,we can find out that class C 1 expands its class size updated by AdaC2.M1 more than thatupdated by AdaBoost.M1.
This observation shows that we can use the cost values to adjust the data distributions among classes. For those classes with poor performances, we can associate them with relative higher cost values such that relatively more weights are accumulated on those parts. As a result,learn-ing will bias and more relevant samples might be identi¬fied. However,if weights are over boosted,more irrelevant samples can be included simultaneously. Precision values of these classes and recall values of the other classes will be decreased. Hence, how to figure out an efficient cost
Proceedings of the Sixth International Conference on Data Mining (ICDM'06)
0-7695-2701-9/06 $20.00 © 2006
k 1 i/k
G — mean — Ra I
a-1 J
A Real-Time Image Stabilization System Based on
Fourier-Mellin Transform
J.R. Martinez-de Dios and A. Ollero
Grupo de Robótica, Visión y Control.
Departamento de Ingenieria de Sistemas y Automatica.
Escuela Superior de Ingenieros. Universidad de Sevilla
Camino de los Descubrimientos, sn, 41092, Sevilla (Spain)
Phone: +34 954487357; Fax: +34 954487340
{jdedios, aollero}@cartuja.us.es
Abstract. The paper presents a robust real-time image stabilization system based on the Fourier-Mellin transform. The system is capable of performing image capture-stabilization-display at a rate of standard video on a general Pentium III at 800 MHz without any specialized hardware and the use of any particular software platforms. This paper describes the theoretical basis of the image matching used and the practical aspects considered to increase its ro-bustness and accuracy as well as the optimizations carried out for its real-time implementation. The system has been submitted to extensive practical experi-mentation in several applications showing high robustness.
1 Introduction
Image stabilization is a relevant topic in many applications including those in which the video is only used for human visualization and those in which the sequences of images are processed by a computer. In human visualization applications image vi¬brations introduce stress in the operator, which involves a decrease in the capacity of attention. In computerized image processing applications vibrations have harmful effects and they often include a step devoted to vibrations cancellation.
Two main approaches have been developed for image stabilization. The first one aims to stabilize the camera vibrations. This approach is used by various types of systems from simple mechanical systems for handheld camcorders to inertial gyrosta-bilized camera systems for gimbals. Mechanical systems for handheld camcorders usually have low accuracy and perform “vibrations reduction” more than “vibrations cancellation”. Gyrostabilized camera systems are restricted to only some applications to due their usual high cost, size and weight. Another approach corrects the images by applying image processing techniques. Several image-processing stabilization meth¬ods have been proposed. The main limitation of these methods is that they require high time-consuming computations. This paper presents a robust real-time image
A. Campilho, M. Kamel (Eds.): ICIAR 2004, LNCS 3211, pp. 376–383, 2004.
© Springer-Verlag Berlin Heidelberg 2004
A Real-Time Image Stabilization System Based on Fourier-Mellin Transform 377
stabilization system based on Fourier-Mellin matching. One of its objectives is to avoid any dependence with hardware or software platforms.
2 Principle for Image Stabilization
The scheme of the processing applied is depicted in Fig.1. Image Stabilization per¬forms image matching techniques between the current image, Imn (x,y), and the
stabilized version of the image captured in the previous time instant, ImE ( x,y)
n 1 .
Image stabilization corrects Imn (x,y) by applying geometric relations inverse to
those found between Imn (x,y) and ImE ( x,y)
n 1 .
Fig. 1. Operation scheme of the application of image stabilization.
Preliminary simulations showed that image vibrations can be modeled as the com¬bination of image translations and rotations while the scale component could be ne¬glected. The most common approach to image matching is based on cross-correlation, [2]. The straightforward formulation of cross-correlation is not capable of matching rotated images. Besides, its has poor selectivity capacity and does not behave well in presence of noise. Some alternatives have been proposed to cope with rotated images. But, these alternatives involve very time-consuming computations. Matching tech¬niques based on invariant moments are sensitive to noise and have low discriminating capacity, [1]. Other group of techniques is based on matching a number of features present in both images [5]. These techniques require the presence of a considerable number of features which could not always be present in the images. Moreover, the features matching is usually carried out by cross-correlation, with the above-mentioned limitations. Several Fourier transform based matching techniques have been proposed in [7] and [6] but they can not match rotated images. Fourier-Mellin transform is capable of matching translated and rotated images, [3].
378 J.R. Martinez-de Dios and A. Ollero
2.1 Matching of Two Images Through the Fourier-Mellin Transform
Consider that image Imnx,y is a rotated and translated replica of image ImEx,y
n .
The stabilization consists of two steps: rotation correction and translation correction. Consider that sx,y and rx,y are respectively the central rectangular region of Imnx,y and ImEx,y . Thus, sx,y is a rotated and translated replica of
n 1
rx,y :
sx,yrxcos v sin x0, x sin v cos y0, (1)
where is the rotation angle and x0 and y0 are the translational offsets. The Fourier transforms of sx,y and rx,y are related by:
S u,v e j s u,v R ucos vsin , usin vcos
, (2)
where su,v is the spectral phase of sx,y and depends on the translation and rotation. The spectral magnitude of Su,v is translation invariant:
Su,v Rucosvsin , usinvcos (3)
Thus, a rotation of the image involves a rotation of the same angle of the spectral magnitude. Assume that rp, and s p, are the spectral magnitudes of rx,yand sx,y in the polar co-ordinates , . It is easy to check that:
sp,rp, (4)
The image rotation is transformed to a translation along the angular axis. It is easy to observe that S p, and R p,, the Fourier transform of s p, and rp,, are related by S p, e R p ,
j2 . Thus, s p, and rp,
have the same spectral magnitude.
2.2 Image Rotation Correction
From ,
Sp and ,
Rp , phases of S p, and Rp,, we define: exp j
r (5)
Q,
Sp
,Rp,
The rotation angle between sx,y and rx,y can be obtained by computing the
inverse Fourier transform of Qr,, qr, F1Qr,, where 1
F stands
for the inverse Fourier transform. The peak of qr, is located at max and
max. The rotation angle is given by max. Once the rotation angle has
A Real-Time Image Stabilization System Based on Fourier-Mellin Transform 379
been obtained the translation correction is applied by rotating Imnx,y an angle . The rotation-corrected image is ImRx,y
n , its central part is sRx,y.
2.3 Image Translation Correction
Qu,v exp j
t
R u,v R u,v
S
The translations between sRx,y and rx,y can be obtained by computing the inverse Fourier transform of Qtu,v, qtx,y Fourier1Qtu,v. The peak of
qtx,y is located at x xmax and y ymax. The translations between sRx,yand rx,y are given by x0 xmax and y 0 ymax . The translation correction is
applied by shifting ImR x,y
n by x0 and y0.
3 Practical Aspects
3.1 Drift Correction
The accumulation of small errors along the stabilization of a certain number of im
ages produces. A drift correction technique is periodically carried out every N of
images. The rotation angle between Imnx,y and ImEx,y
n 1 is 1
and, between
Imnx,y and ImEx,y
nN is N
. The rotation angle is taken as a combination of
them:
1r1rN, (7)
where r0, 1 is called drift correction rotation factor. If r 0 no drift correc¬tion is applied, while r1 usually generates sudden changes in image rotation cor¬rections. Once Imnx,y has been rotation corrected with , the translational off¬sets are computed by combining x0 1 and y0 1, the translational offsets between ImR x,y
n and ImEx,y
n 1 , together with N
x0 and N
y0 , translational offsets
between ImR x,y
n and ImEx,y
nN . The expressions of such combinations are:
380 J.R. Martinez-de Dios and A. Ollero
x0 1 = (1 τt ) x0 1 + τt x0 N and y0 1 = (1 τt ) y0 1 + τt y0 N , where τt [0, 1] is the drift translation correction factor.
3.2 Contrast Correction
The matching method obtains poorer results with low contrasted images. A contrast-enhance method based of histogram stretching has been used to improve the lumi¬nance of the images. Luminance of Im(x,y) is often characterized by the image
bright (MI) and contrast (C). The transformation function that should be applied to obtain the desired bright and contrast values (MIref and Cref ) is:
Im*(x,y)=(Cref C)(Im(x,y)MI)+MIref .
3.3 Operation Modes
The resolution in the computation of the rotation angle and translational offsets is highly dependent on the size of the matrices that represent the images. Consider that r(x,y) and s(x,y) are represented by square matrices of size MxM and that sp (θ, ρ)
and rp (θ, ρ) are represented by square matrices of size WxW.
The resolution of the rotation angle depends on the size of the matrices that repre¬sent Sp (ν,ϖ) and R p (ν,ϖ), i.e. WxW. It is easy to notice that the minimum detect
able angle is αmin 2 W. The value of W also has influence on the errors in the
computation of the rotation angle. The higher W is, the more accuracy in the compu¬tation of the rotation angle can be obtained. The value of W depends on the number of different radius values considered in the polar conversion, which is constrained by the size of S(u,v) and R(u,v) . The size of the matrices that represent r(x,y) and
s(x,y) has straightforward influence on the resolution and errors in the computation
of the translations. Lower values of M involve poor accuracy since the peak of qt (x,y) is more broad and more affected by noise. Two operation modes have been
selected to cope with the compromise between computer requirements and stabiliza¬tion accuracy: Mode1 (low values of M and W and medium stabilization capability) and Mode2 (for high magnitude or high frequency vibrations).
3.4 Increase Accuracy Through Sub-pixel Resolution
The position of the peak of qr (θ, ρ) and qt (x,y) determine the value of α , and x0 and y0, respectively. The resolutions in the computation of α , x0 and y0 are lim
ited by the values of M and W. Selecting higher values of M and W increases the computation load. An efficient alternative applied considers a sub-pixel estimation to the peak position. The sub-pixel estimation considers the peak is located at the cen-troid in a neighborhood of certain size centered at the position of the peak:
A Real-Time Image Stabilization System Based on Fourier-Mellin Transform 381
p_m AX 1B X CX 1A B C, where A, B and C are the magnitude
at X-1, X and X +1 respectively. For estimating the position of the peak in matrices, the centroid is computed in a 2D neighborhood.
Fig. 2. Sub-pixel estimation of peak position.
4 Computational Aspects
Image matching based on Fourier Mellin requires the computation of six 2D FFTs and two 2D Inverse FFTs. Special care has been put in the optimization of the com¬putation of the FFT. The Cooley-Tukey FFT algorithm [4] was used due to its combi¬nation of efficiency and simplicity. The size of the matrices has been selected to be power of two. The row-column approach was used for the 2D FFT.
Two approaches have been considered for the optimization of the Fourier trans¬forms. The first one exploits the symmetry properties of FFTs of real data. If xk is
a sequence of real data its FFT, X k, satisfies: ReX k ReX k and ImX kImX k, where ReX k and ImX k are the real and imaginary components of X k. The computation of the 2D FFT also exploits the following
symmetry property: if A(x,y) is a matrix of real data its 2D FFT, A(u,v), satisfies: ReAu,vReAu, v and ImAu,vImAu, v. Both properties can
save up to 50% of the total computation of the 2D FFT of matrices with real data.
The second one computes the twiddle factors of the Cooley-Tukey algorithm at the initialization of the stabilization system. The twiddle factors, Wi e2
i N, are con
stant values that only depend on N, the length of the vectors which FFT is to be com¬puted, which only depend of the operating mode. The pre-computation of the twiddle factors avoids calculating them each time a FFT is computed. This represents an im¬portant reduction (more than 40%) in the operations required for FFT.
Rotation and translation corrections involve the application of image interpolation to deal with non-integer rotation angles and translational offsets. Bilinear interpola¬tion was chosen for its simplicity and efficiency. Further reduction in the computa¬tional load (up to 30%.) of bilinear interpolations can be obtained by using ‘integer’ Mathematical operations which are more efficient than ‘floating point’.
382 J.R. Martinez-de Dios and A. Ollero
5 Experiments
The image stabilization method was implemented with ANSI C on a Pentium III at 800 MHz. It was implemented in Windows NT and Vx-Works to test its portability. The system was submitted to extensive experiments in several different applications. Consider that ImE x,y and Imn x,y , shown in Fig. 3a-b, are two consecutive
n 1
images. The first step is the computation of the rotation angle between ImEx,y
n 1
and Imnx,y. The peak of qr, takes place at max 3, which corresponds to 2'109375 º. Then, Imnx,y is rotated an angle º. The rotation corrected image, ImRx,y
n , is shown in Fig. 3c. In the computation of the translational offsets,
the peak of qtx,y takes place at x0 =0.4 and y0 =-585. Translation correction is applied by shifting ImR x,y
n - x0 and - y0. The stabilized image ImE x,y
n is
shown in Fig. 4c together with the original image (shown in Fig. 4b) and reference image, in Fig. 4a.
a) b) c)
Fig. 3. a), b) Two consecutive images from a camera under vibrations ImE x,y and
n 1
Imnx,y, c) the rotation-corrected version of Imnx,y.
Numerous experiments have been carried out to test the robustness of the system. In the experiments carried out Mode1 uses M=W=128 and Mode2, M=W=256. The image stabilizing time at Mode1 is 28.6 ms., which allows real-time stabilization for PAL and NTSC video standards. The stabilizing time at Mode2 is 102.1 ms.
a) b) c)
Fig. 4. a) ImEx, y ; b) original image with vibrations Imnx, y ; c) stabilized ImEx, y.
n 1 n
A Real-Time Image Stabilization System Based on Fourier-Mellin Transform 383
6 Conclusions
The paper presents a robust real-time image stabilization system based on Fourier-Mellin transform. The stabilization system is based on applying matching based on Fourier-Mellin transforms between consecutive images in a sequence. The stabiliza¬tion system was optimized to correct the rotations and translations since the scale factor between consecutive images could be neglected in the applications considered. Image matching is applied in two steps: detection and correction of rotations and detection and correction of translations. To increase the robustness, the system in¬cludes drift correction techniques and contrast correction. To increase the accuracy of the system, it includes sub-pixel computation of the rotation angle and translational offsets. Special effort has been applied on the minimization of the computer load including -among others- the pre-computing of Cooley-Tukey twiddle factors and the computation of several operations with integer data. The method was implanted in a Pentium III at 800 MHz with 128 Mbytes of RAM. It is capable of performing image capture-stabilization-display at a rate of PAL and NTSC video.
Acknowledgements. The authors would like to thank Joaquin Ferruz and Luis Fernández. The work described in this paper has been developed in the project SEOAN “Sistema Electroóptico de Ayuda a la Navegación”. SEOAN project is leaded by “Division de Sistemas” of the Spanish company IZAR and funded by the “Gerencia del Sector Naval”. The authors express their gratefulness to Antonio Criado, Francisco López, Alfonso Cardona, Baltasar Cabrera, Juan Manuel Galán and José Manjón from IZAR.
References
1. Abu-Mostafa Y.S. and D. Psaltis. “Recognition aspects of moment invariants”. IEEE Trans. Pattern Anal. Mach. Intel., 16(12) (1984). 1156-1168.
2. Barnea, D. I. and H. F. Silverman, “A class of algorithms for fast image registration”. IEEE Trans. Computers, C-21, (1972). 179-186.
3. Chen Q., M. Defrise, F. Deconinck, “Symmetric Phase-Only Matched Filtering of Fourier-Mellin Transform for Image Registration and Recognition”, IEEE Trans. P.A.M.I., vol. 16, no 12, (1994). 1156-1167.
4. Cooley J.W. and J.W. Tukey, “An algorithm for the machine calculation of complex Fourier series”, Math. Comput. 19, (1965). 297–301.
5. Faugeras O., Q. Luong, and T. Papadopoulo, "The Geometry of Multiple Images. MIT Press, 2001. ISBN 0-262-06220-8.
6. Horner J.L. and P.D. Gianino, “Phase-only matched filtering”, Applied Optics, vol. 23, no. 6, (1984). 812-816.
7. Oppenheim A.V. and J.S. Lim, “The importance of phase in signals”, IEEE Proc. Vol. 69, no. 5, (1981). 529-541.
Graph Pattern Spaces from Laplacian Spectral Polynomials
Bin Luo, Richard C. Wilson, and Edwin R. Hancock
Department of Computer Science, University of York, York Y010 5DD, UK.
Abstract. Graph structures have proved computationally cumbersome for pattern analysis. The reason for this is that before graphs can be converted to pattern vec¬tors, correspondences must be established between the nodes of structures which are potentially of different size. To overcome this problem, in this paper we turn to the spectral decomposition of the Laplacian matrix. We show how the elements of the spectral matrix for the Laplacian can be used to construct symmetric polyno¬mials that are permutation invariants. The co-efficients of these polynomials can be used as graph-features which can be encoded in a vectorial manner. We explore whether the vectors of invariants can be embedded in a low dimensional space using a number of alternative strategies including principal components analysis (PCA), multidimensional scaling (MDS) and locality preserving projection (LPP).
1 Introduction
The analysis of relational patterns, or graphs, has proved to be considerably more elusive than the analysis of vectorial patterns. One of the challenges that arises in these domains is that of knowledge discovery from large graph datasets. The tools that are required in this endeavour are robust algorithms that can be used to organise, query and navigate large sets of graphs. In particular, the graphs need to be embedded in a pattern space so that similar structures are close together and dissimilar ones are far apart. Moreover, if the graphs can be embedded on a manifold in a pattern space, then the modes of shape variation can be explored by traversing the manifold in a systematic way. The process of constructing low dimensional spaces or manifolds is a routine procedure with pattern-vectors. A variety of well established techniques such as principal components analysis, multidimensional scaling and independent components analysis, together with more recently developed ones such as locally linear embedding [5], isomap [6] and locality preserving projection [7] exist for solving the problem. However, there are few analogous methods which can be used to construct low dimensional pattern spaces or manifolds for sets of graphs.
There are two reasons why pattern-vectors are more easily manipulated than graphs. First, there is no canonical ordering for the nodes in a graph, unlike the components of a vector. Hence, correspondences to a reference structure must be established as a prerequisite. The second problem is that the variation in the graphs of a particular class may manifest itself as subtle changes in structure, which may involve different numbers of nodes or different edge structure. Even if the nodes or the edges of a graph could be encoded in a vectorial manner, then the vectors would be of variable length. One
A. Campilho, M. Kamel (Eds.): ICIAR 2004, LNCS 3211, pp. 327–334, 2004.
c Springer-Verlag Berlin Heidelberg 2004
328 B. Luo, R.C. Wilson, and E.R. Hancock
way of circumventing these problems is to develop graph-clustering methods in which an explicit class archetype and correspondences with it are maintained [9]. One of the problems that hinders graph clustering is the need to define a class archetype that can capture both the salient structure of the class and the modes of variation contained within it. For instance, the random graphs of Wong, Constant and You [11] capture this distribution using a discretely defined probability distribution, and Bagdanov and Worring [10] have overcome some of the computational difficulties associated with this method by using continuous Gaussian distributions.
However, one of the criticisms that can be aimed at these methods for learning the distribution of graphs is that they are in a sense brute force because of their need for correspondences either to establish an archetype or to compute graph similarity. For noisy graphs (those which are subject to structural differences) this problem is thought to be NP-hard. Although relatively robust approximate methods exist for computing correspondence [1], these can prove time consuming. In this paper, we are interested in the problem of constructing pattern spaces for sets of graphs. To do this we vectorise graphs and use standard manifold learning methods to project the vectors into a metric space. Our aim is to construct the vectors in a way which does not require the computation of node correspondences. We hence turn to a spectral characterisation of graph-structure [2]. Although existing graph-spectral methods have proved effective for graph-matching [3] and indexing [8], they have not made full use of the available spectral representation, and are restricted to the use of either the spectrum of eigenvalues or a single eigenvector. Moreover, in prior work we have demonstrated that simple hand-crafted spectral features that reflect the “shape” of graphs can be used for clustering [4]. The aims here are more ambitious since our characterisation is based on permutation invariants computed from the elements of the full spectral matrix. The representation is hence both more principled and richer.
2 Graph Spectra
In this paper we are concerned with the set of graphs G1, G2,.., Gk, ..., GN. The kth graph is denoted by Gk = (Vk, Ek), where Vk is the set of nodes and Ek Vk × Vk is the edge-set. These graphs are undirected. Our approach in this paper is a graph-spectral one. For each graph Gk we compute the adjacency matrix Ak. This is a |Vk| × |Vk| matrix whose element with row index i and column index j is
1 if (i, j) Ek
Ak(i, j) =
(1)
0 otherwise
The aim in this paper is to perform spectral analysis. Unfortunately, the adjacency matrix can have negative eigenvalues. Instead we turn our attention to the Laplacian matrix, since it is positive semi-definite and therefore has positive or zero eigenvalues. For the graph indexed k the Laplacian matrix is Lk = Ak Dk where Dk is diagonal degree matrix with elements of Dk(i, i) = EjVk Ak(i, j). The spectral decomposi¬tion of the Laplacian matrix is Lk = ni=1 λk i ek i ek T , where λk
i i is the ith eigenvalue
and eki is the corresponding eigenvector of the Laplacian matrix Lk of the graph in¬dexed k. With the eigenvalues and eigenvectors at hand the spectral matrix is given by
Graph Pattern Spaces from Laplacian Spectral Polynomials 329
Φk = λk1 ek1 λk2 ek2 ... λn ek|Vk| . Since the eigevalues of the Laplacian matrix
are real and positive or zero, the corresponding roots are real. The graph spectral matrix Φk is a complete representation of a graph Gk in the sense that the original Laplacian matrix which depicts the adjacency of graph nodes can be reconstructed using the rela¬tionship Lk = ΦkΦTk . Moreover, Φk is unique for a simple graph G (without repeated eigenvalues) up to a permutation of columns. Hence, we can not use the elements of Φk to construct useful graph features, since they are not permutation invariants. To over¬come this problem we use symmetric polynomials to compute permutation invariants. The polynomials are defined as follows
S1(x1,...xn) =
...
Sr(x1,...xn) = xi1xi2 ... xir
i1<i2<...<ir
...
Sn(x1, ... xn) =
To overcome the problem that the resulting distributions may be extremely long-tailed, we take logarithm of the elements of S as the new features,
Fki,j = signum(Ski,j) ln(1 + Ski,j ) (2)
We form a feature vector Bk for graph Gk using the standard vectorisation operation on the elements of the matrix F
Bk=(Fk1,1, ... ,Fk1,n,Fk2,1, ... ,F k2,n,...,Fkn,1,...,Fkn,n)T. (3)
In order to accommodate graphs of different sizes, we need to be able to compare representations of different sizes. This is achieved by expanding the representation. Consider two graphs of size m and n, m < n. If we add nm nodes with no connections to the first graph, we obtain two graphs of the same size. The edit cost in terms of edge insertions and deletions between these two graphs is identical to the original pair. The effect on the spectral representation is merely to add trailing zeros to each eigenvector and additional zero eigenmodes. As a consequence, the first m elementary symmetric polynomials are unchanged, and the subsequent n m are zero. The new representation in S can therefore be easily calculated from the original feature vector.
3 Graph Embedding Methods
We explore three different methods for embedding the graph feature vectors in a pattern space.
330 B. Luo, R.C. Wilson, and E.R. Hancock
Fig. 1. Examples images and corresponding graphs.
3.1 Principal Component Analysis
We commence by constructing the matrix S = [B1|B2| ... |Bk| ... |BN ]. with the graph feature vectors as columns. Next, we compute the covariance matrix for the elements of the feature vectors by taking the matrix product C = SST.. We extract the princi¬pal components directions by performing the eigendecomposition C = ~Ni=1 liuiuTi on the covariance matrix C, where the li are the eigenvalues and the ui are the eigen-vectors. We use the first s leading eigenvectors ( 2 or 3 in practice for visualisation purposes) to represent the graphs extracted from the images. The co-ordinate system of the eigenspace is spanned by the three orthogonal vectors Ue = (u1, u2, .., us). The individual graphs represented by the long vectors Bk, k = 1, 2, ... , N can be pro¬jected onto this eigenspace using the formula xk = UTBk.. Hence each graph Gk is represented by an s-component vector xk in the eigenspace.
3.2 Multidimensional Scaling
Multidimensional scaling(MDS) is a procedure which allows data specified in terms of a matrix of pairwise distances to be embedded in a Euclidean space. Here we intend to use the method to embed the graphs extracted from different viewpoints in a low-dimensional space. To commence we require pairwise distances between graphs. We do this by computing the L2 norms between the spectral pattern vectors for the graphs. For the graphs indexed i1 and i2, the distance is di1,i2 = (Bi1 Bi2)T (Bi1 Bi2). In this paper, we use the classical multidimensional scaling method to embed the view-graphs in a Euclidean space using the matrix of pairwise dissimilarities D. The first step of MDS is to calculate a matrix T whose element with row r and column c is given by
Trc = 21 [d2rc ˆd2r. ˆd2.c + ˆd2 ..], where ˆdr. = 1 ~N c=1 drc is the average dissimilarity
N
Graph Pattern Spaces from Laplacian Spectral Polynomials 331
value over the rth row, EN ˆd.c is the similarly defined average value over the cth column and
ˆd.. = 1 EN
N2 r=1 c=1 dr,c. is the average dissimilarity value over all rows and columns
of the matrix T.
We subject the matrix T to an eigenvector analysis to obtain a matrix of embedding co-ordinates X. If the rank of T is k, k N, then we will have k non-zero eigenvalues. We arrange these k non-zero eigenvalues in descending order, i.e. l1 >_ l2 >_ ... >_ lk > 0. The corresponding ordered eigenvectors are denoted by ui where li is the ith eigenvalue. The embedding co-ordinate system for the graphs obtained from different views is X = [f1, f2, ... , fs], where f i = liui are the scaled eigenvectors. For the graph indexed i, the embedded vector of co-ordinates is xi = (Xi,1, Xi,2, ..., Xi,s)T ..
3.3 Locality Preserving Projection
Our next pattern space embedding method is He and Niyogi’s Locality Preserving Pro-jections(LPP) [7]. LPP is a linear dimensionality reduction method which attempts to project high dimensional data to a low dimensional manifold, while preserving the neigh¬bourhood structure of the data set. The method is relatively insensitive to outliers and noise. This is an important feature in our graph clustering task since the outliers are usually introduced by imperfect segmentation processes. The linearity of the method makes it computationally efficient.
The relational structure of the data is represented by a proximity weight matrix W with elements W (i1, i2) = exp[kd(i2, i2)], where k is a constant. If Q is the diagonal degree matrix with the row weights Q(k, k) = ENj=1 W(k, j) as elements, then the relational structure of the data is represented using the Laplacian matrix J = QW. The idea behind LPP is to analyse the structure of the weighted covariance matrix SW ST . The optimal projection of the data is found by solving the generalised eigenvector problem SJST u = lSQST u.. We project the data onto the space spanned by the eigenvectors corresponding to the s smallest eigenvalues.
4 Experiments
Our experimental vehicle is provided by 2D views of 3D objects. We have collected sequences of views for a number of objects. For the different objects the image sequences are obtained under slowly varying changes in viewer angle. From each image in each view sequence, we extract corner features. We use the extracted corner points to construct Delaunay graphs. In our experiments we use three different sequences. Each sequence contains images with equally spaced viewing directions. For each sequence we show a sample of the images and the resulting graph structure in Figure 1.
In Figure 2 we compare the results obtained with the different embedding strategies and different graph features. In the left-hand column, we show the results obtained when PCA is used, the middle column when MDS is used, and the right-hand column when LPP is used. The top row show the results obtained using a standard spectral feature, namely the spectrum ordered eigenvalues of the Laplacian i.e. Bk = (λk1, λk2, ...)T . The second row shows the embedding obtained by computing the symmetric polynomials, using the spectral matrix for the graph adjacency matrices. The third row shows the results
334 B. Luo, R.C. Wilson, and E.R. Hancock
References
1. W. J. Christmas, J. Kittler and M. Petrou, “Structural Matching in Computer Vision using Probabilistic Relaxation”, IEEE PAMI, 17, pp. 749–764, 1995.
2. F.R.K. Chung, “Spectral Graph Theory”, American Mathmatical Society Ed., CBMS series 92, 1997.
3. S. Umeyama, “An eigen decomposition approach to weighted graph matching problems”, IEEE PAMI, 10, pp. 695–703, 1988.
4. B. Luo, R. C. Wilson and E. R. Hancock, “Spectral Embedding of Graphs”, Pattern Recog¬nition, 36, pp. 2213–2230, 2003
5. S. Roweis and L.Saul, “Non-linear dimensionality reduction by locally linear embedding”, Science, 299, pp. 2323-2326, 2002.
6. J.B. Tenenbaum, V.D. Silva and J.C.Langford, “A global geometric framework for non-linear dimensionality reduction”, Science, 290, pp. 586–591, 2000.
7. X.He and P. Niyogi, “Locality preserving projections”, to appear in NIPS03.
8. A. Shokoufandeh, S. Dickinson K. Siddiqi and S. Zucker, “Indexing using a Spectral Coding of Topological Structure”, CVPR, pp. 491–497, 1999.
9. J. Segen, “Learning graph models of shape”, in J. Laird, editor, Proceedings of the Fifth International Conference on Machine Learning, pp. 29–25, 1988.
10. A.D. Bagdanov and M. Worring, “First Order Gaussian Graphs for Efficient Structure Clas¬sification”, Pattern Recogntion, 36, pp. 1311-1324, 2003.
11. A.K.C Wong, J. Constant and M.L. You, “Random Graphs”, Syntactic and Structural Pattern Recognition, World Scientific, 1990.
QUALITY ASSURANCE OF LIDAR SYSTEMS –
MISSION PLANNING
Kutalmis Saylam
GeoBC
Crown Registry and Geographic Base (CRGB) Branch
1st Floor, 3400 Davidson Ave,
Victoria, BC V8Z 3P8
Canada
Kutalmis.saylam@gov.bc.ca
ABSTRACT
Mission planning is considered a crucial aspect of Airborne Light detection and Ranging (LiDAR) surveys to contribute to total Quality Assurance (QA) experience. Since LiDAR is a relatively new spatial data acquisition practice, one may not find complete documentation on how to get prepared for such a mission. There is abstract information available from a few public and private organizations; however, none of these resources provide fully documented and thorough explanation. Throughout the industry, most airborne LiDAR missions are prepared with the previous expertise of the personnel who have involved in earlier projects. Formal training is not common, and ‘learning-on-the-job’ has potential complications for the future. Additionally, there are various types of airborne LiDAR surveys that require specific know-how, but expertise carried over may not work for a different type of survey. Field and office managers are advised to assess the project requirements and available resources very carefully prior to mission initiates. There are fundamental requirements, as well as less important actions. Due to the variable nature of airborne surveys, all phases need steady observation to prevent potentially costly changes or mission failures. Various projects experience difficulties in order to complete projects sooner, resulting in overlooked and skipped QA procedures. Careful assessment of the requirements and planning with adequate timing is vital for a successful mission completion. A good mission planning requires careful and extensive consideration of various phases of the project. Hence, author believes that there is a need for detailed airborne LiDAR mission planning documentation that will provide assistance to LiDAR community.
INTRODUCTION
Quality Assurance procedures refer to planned and systematic processes that provide confidence of a product’s or service’s effectiveness. This applies to all forms of activities; design, development, production, installation, servicing and documentation stages.
QA for airborne LiDAR mission planning refers to forecasting and management activities to ensure that proposed mission is executed and completed with highest possible quality that is available. These activities in general would include good mission planning, accurate system configuration, well documented data processing and complete project delivery. Figure 1 illustrates a general QA model flowchart.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Figure 1. Quality Assurance Model for a LiDAR Mission.
ESSENTIALS OF AIRBORNE MISSION PLANNING
Accurate flight planning for airborne LiDAR survey is essential for a total Quality Assurance experience. Due to the variable and challenging nature of airborne surveys, there are various issues that should be assessed properly. Many people involved at this stage of a LiDAR mission have on-the-job expertise, and do not possess formal training about LiDAR data acquisition basics. Since expenses related to a flight mission are very high, proper mission planning at this stage is very important. Figure 2 illustrates activities towards reaching data acquisition stage.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Figure 2. Mission Planning Activities.
Some of the planning essentials and recommendations for airborne LiDAR survey mission are as follows:
1. Initial review of survey area is essential for effective survey planning. Existing imagery or maps may be
used for a complete assessment.
- Local weather patterns, sudden topography changes, existing water basins, and general terrain cover may impact flight planning and should properly be examined prior to flight planning.
- Large survey areas (>200 km2) need multiple flight plans (e.g. 4 x 50 km2) that should accommodate 20-30% overlapping to prevent loss of data in case of a system failure/reset and easier data processing/handling
- Large survey areas with multiple segments should have flight plans prepared for each survey day and each segment should have sufficient overlapping
- Borders of a flight plan should exceed the physical borders of the survey area with additional cross strips planned at the end of a survey mission for QC process (system calibration)
- Multiple flight plans should be prepared to accommodate different flight altitudes; various factors may have an impact on the flight mission such as changing cloud cover, weather turbulence, flight altitude restrictions, and etc.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
- The flight plan parameters should have enough tolerance for variable factors; such as if 1 point is required for each m2 per contract; flight plan should accommodate up to 1.1 to 1.5 points per m2 (10% to 50% tolerance)
2. Selection of aircraft type; either fixed wing or a helicopter
- Fixed wing aircrafts are more suitable for large area surveys, while helicopters provide better functionality for corridor survey types, however they are more expensive to operate
3. If project outcome requires Digital Elevation Model (DEM) creation, point density needs to be adjusted accordingly through flight plans. Rule of thumb is to collect data with 1 metre point spacing for each 1/2 metre contour interval (e.g. 4 metre point spacing is required to create 2 metre contour interval)
4. Selection and planning of base station use for post processing GPS data; a single standalone GPS, Virtual Reference Network (VRS) or Precise Point Positioning (PPP)
- If a single station scenario is planned for the survey, flight plan borders should not exceed the recommended distance (30 km radius) to maintain good GPS accuracy
- 2 base stations should be available for a single station scenario for backup purposes, and both should run simultaneously during a mission.
- Base stations should have capability to collect data with an ellipsoidal height average accuracy of 0.02 metres or better, and have dual-frequency (L1/L2) capability with plenty of internal data storage capacity
- Position Dilution of Precision (PDOP) values must be checked well in advance and properly examined for survey mission dates. High PDOP values will affect the final product accuracy adversely.
5. All parameters and configuration settings such as GPS antenna offset measurements, communication protocols, computer to LiDAR system connections, and etc should be checked before mission commences to prevent any mission failures or delays.
GPS Antenna Offset Measurement
This procedure is required after a LiDAR system is installed in an aircraft for the first time. There is no need to repeat the procedure on condition that there have been no position changes to GPS antenna, the mounted sensor head, and the aircraft sensor head opening port.
POS (Position and Orientation System) computer needs to register the exact location of GPS antenna phase center relative to the scanner mirror. Therefore, a precise measurement between reference point on the sensor to the phase center of the GPS antenna with using a Total Station and a reflective prism is required. To comply with LiDAR Quality Assurance procedures, this measurement is vital for accuracy of the final product. Precision of such a measurement is expected to be better than 0.02 metres.
Components measured to fore of the point of origin (to nose of the aircraft) has +X polarity, while measurements to aft of the point of origin has the opposite; -X polarity. Components measured to right has +Y, while measurements to left has –Y polarity. Any components measured below the point of origin have + Z and components above are referred with –Z (See above Figure 3).
X, Y, Z offset distance from the scanner mirror to the reference point is always measured at the manufacturer’s lab before sensor head housing is sealed. These internal values are provided to the client when a new system is delivered. Each system manufactured may have different values. Values measured with Total Station from the reference point to the phase center of the GPS antenna (as explained above) are added to the internal lab values. The final result illustrates the relative X, Y and Z distances between the scanner mirror and the GPS antenna phase center. These values are entered into Position Orientation system application (e.g. POS/AV Controller) for offset measurement corrections.
Control Surface and Targets Preparation
For any spatial data acquisition system, it is essential to ensure that Quality Assurance procedures are completed and the final product meets the end users’ needs and accuracy requirements1. Since a typical LiDAR survey includes data acquired in many overlapping strips, a common QA procedure is to correctly establish a control surface that will assist to analyze the coincidence of overlapping laser points collected in opposite directions. The degree of coincidence of points is used to identify the presence of systematic biases in the LiDAR system. Common methods with using control surfaces and distinctive terrain features are explained very briefly in next chapters.
Control Surface – Long & Flat Surface. A ground control surface is required only to determine absolute accuracy. Control points on the surface are measured with a high precision survey instrument (Geodetic level GPS or Total Station) which is tied into local control network with a first order survey monument, horizontally and
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
vertically. It is expected that control surface points would have at least 3 times better accuracy than airborne laser scanner points. Yet, absolute accuracy analysis is only necessary to calibrate the LiDAR system. LiDAR system manufacturers advise to repeat absolute calibration every 6 months or so, even if the system produces good accuracy values.
To create an ideal control surface, a long (1500 to 2000 metres), wide (30 to 50 metres) and a linear surface is required. Control points should be in 5 – 10 metres grid, and possess very little blockage to prevent measurements. Ideally, this would be a municipal airport runway, or a long , straight non-residential stretch of a rural highway. If an airport is preffered, it is advised to check with airport authorities and inquire about flight restrictions in the area. Usually Class D & E airspaces are suitable for such calibration purposes.
A survey grade GPS base station (dual frequency L1/L2 preferred) needs to be established in the close proximity of the calibration site (< 30 kms). GPS base station collects the data that is required for post processing. Ideally, GPS receiver should be capable of data logging at 1 Hz. Additionaly, if possible, a second base station should be running simultaneously for backup purposes. If there are VRS or CORS networks in the area, this should also be considered.
Other Control Features – Buildings, Trees, etc... A ground control feature can be used for both absolute and relative accuracy. Large buildings (80 to 200 metres wide) are preferred since it is difficult to hit the building at an altitude of 1000 metres or more at profile mode (Scanner is set to 0º) to estimate the pitch bias. Also it is important that buildings are higher than 5 metres with no unusual overhangs, ledges and long banks of windows that might catch or reflect laser points. Typically, control points are located at the edge of the corners of the building.
If there are no control features located on the survey location, other terrain features may be used for relative accuracy assessments. These features would be tall trees, smaller buildings, bridges and etc. Investigation of the degree of coincidence of laser points with overlapping laser strips in opposite directions over these features would enable finding of systematic errors present in the LiDAR system.
LiDAR Specific Control Features. There are applications that require higher accuracy, such as transportation corridor mapping and power lines survey. Building and using well-defined and accurate control targets shall contribute to the success of Quality Assurance procedures and the final product. The objective with building a LiDAR-specific ground target is to provide a highly accurate positioning in both horizontal and vertical directions. Csanyi, N and C. K. Toth (2007)2 discuss that due to different possible scan directions and different point densities in different directions, the optimal LiDAR target must be rotation invariant, circle-shaped, and elevated from the ground. Also, targets may have specific coating to provide different reflective surface that can be picked up by distinguished intensity values. After LiDAR survey of targets, the coordinates of extracted targets are compared with laser points using a Root Mean Square Error analysis (RSME).
In general, this practice is considered expensive and time consuming3 since its implementation depends on the accessibility to the survey site. But results indicate that errors larger than 10cm at horizontal and larger than 2-3cm at vertical distances can be detected with this practice. Toth, C. et al (2006)4 confirm 2-3 cm horizontal and 2 cm vertical accuracy.
GPS Data Quality and PDOP Prediction
Positional Dilution of Precision (PDOP): This is the measure of geometrical strength of GPS satellite configuration. Generally, PDOP < 3 is desired but values less than 4 are acceptable for airborne surveys. PDOP 1 is the best value achievable theoretically. With the addition of GLONASS satellites, very low PDOP values are attainable. However, most airborne Positioning systems currently on the market utilize only GPS (USA owned) satellites. In near future, with the further use of additional satellites (GLONASS, Galileo, etc), it is expected that Positioning system manufacturers will integrate full GNSS functionality into their products.
System operator may observe unexpected PDOP spikes during a mission. If spike is considerably high, data acquisition process should be suspended until better GPS geometry is achieved. Since GPS is one of the primary sources of positional error source (alongside Inertial Measurement Unit)5, it is essential to plan ahead for good GPS constellation to comply with QA model. A well-known manufacturer of GPS/Inertial Navigation System specifies 5 to 10 cm for sensor positioning, 0.005º for pitch/roll, and 0.008º for heading error as system tolerance. At a flying altitude of 1000 metre, these orientation errors translate into planimetric errors of 10-15 cm on the ground for each of the three angles6.
PDOP values for the survey location should be predicted well before the mission by field manager, and flight times should be adjusted accordingly. There are programs available to predict the DOP values for specific dates and locations. For instance, System Effectiveness Model for Windows (WSEM36) is a freeware that is available via Internet7. Also, an updated almanac should be downloaded and integrated into WSEM36 before any predictions are completed.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Ground Truthing: Traditional vs. New methods
In the last few years, government agencies and some private organizations began providing post processing GPS data through online services with or without subscription. In Canada; Natural Resources Canada (NRCan) has established continuously operating GPS reference station network referred as Canadian Spatial Reference System (CSRS)8, while some other provincial governments have also established their own services. For instance; in the Province of British Columbia, Crown Registery and Geographic Base Branch9 (CRGB) and in the province of New Brunswick, Service New Brunswick (SNB)10 provides such service to clients. In USA; National Geodetic Survey manages Continuosly Operating Reference Stations (CORS)11 for various locations throughout the nation. Clients have the opportunity to download data with different logging speeds (1Hz – 30 Hz) and with different formats (raw, RINEX).
In the last few years, Virtual Reference Stations (VRS) have become very popular. A VRS solution uses a network of continuously operating reference stations to compute a set of corrections for a roving receiver located anywhere in the network. If survey location has access to any VRS network, there is no need to establish a ground GPS base station. A manufacturer of airborne positioning systems claims to process up to 50 stations at the same time, where the farthest one can be 400 km away12.
Addition to Continuously Operating Reference system, there is another alternative for airborne LiDAR surveys without using an independent GPS base station. The availability of precise GPS satellite orbit and clock corrections has enabled the development of a new positioning method – Precise Point Positioning (PPP).
According to GrafNav13, PPP presents an overall accuracy of 10 to 20 cm. For surveys which do not require very high accuracy levels, these figures are acceptable. For airborne surveys such as corridor mapping which require very long baselines, PPP practice may be an ideal solution.
Important modification with PPP service is to include the replacement of satellite orbits and clock corrections with more precise estimates from an organization such as International GNSS Services (IGS). Rapid satellite position and clock solutions are provided with free of charge from IGS. Data is available usually after 24 hours. IGS stations are well distributed across Canada and USA.
As a Quality Assurance procedure, field manager should ensure the availability, integrity and consistency of ground truth data before a LiDAR mission begins. In a nutshell, quality of GPS data collected during a mission (on the ground and on the air) has a direct impact on the quality of final product.
Cost/Benefit Analysis of a Mission
Mission planning is not complete without a Cost/Benefit Analysis (CBA). Assessment of total expected costs against the total expected gain is an essential aspect of any project. The truth of the outcome of CBA is dependent on how accurately costs and benefits have been estimated. Since LiDAR projects may have several unexpected expenses, a true assessment of ‘required’ and ‘optional’ costs is strongly advised.
Airborne LiDAR data acquisition expenses may depend on many variable factors. Desired point density and size of survey area may be considered as the principal cause for variable costs. Typically, LiDAR system and aircraft leasing costs, crew and flight expenses, data processing and data delivery related fees count as the largest expenses in a mission. Cost would also vary depending on the terrain type and the proximity of the project location to base office. Typically, an airborne LiDAR mission would include expenditures illustrated with Table 1.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Table 1. Typical LiDAR Mission Expenses
Aircraft Leasing and Flight Charges Aircraft leasing (covers insurance & maintenance)
Fuel charges
Airport landing fees
Hangar and ground service fees
Personnel Costs Pilots: Captain and First Officer
LiDAR System Operator
Ground Control Personnel
Travel & Accommodation Costs Air or ground travel for personnel
Lodging and meals at survey location
Data Processing Costs Hardware purchase and maintenance
Software purchase and licensing
Personnel
Data Storage & Handling
Miscellaneous Expenses Ground Truth Surveys/Network Subscription Fees
Calibration flight
Equipment Shipment charges
Other project specific miscellaneous expenses
ACCURATE LiDAR CONFIGURATION
There are various types of airborne LiDAR surveys. Due to diverse nature of airborne surveys, LiDAR systems come equipped with their own mission planning software to configure the laser output. Most surveys require specific parameters depending on the project deliverables. To match the desired ‘Point density’ requirements, LiDAR systems can be configured in various ways. Below is a list of parameters that may be manipulated either before mission or ‘on-the fly’ by the System Operator:
• Scan (speed) frequency (Hz): Different laser scanners have different scanning mechanisms (oscillating, sinusoidal, fiber scanner, rotating polygon, etc.). Most laser scanners currently on the market have moving scanners, and their motion speed is programmable via user interface. This is usually between 0 – 100 Hz with 0.1 Hz increments.
• Scan (width) angle (θ): This applies to swath width and generally referred with half-angle. Typically, half-angle range is between 0 and 30 degrees and can be adjusted with 0.1º increments.
• System Pulse Repetition Frequency (kHz): Laser pulse repetition frequency (PRF) defines number of emitted pulses per second. Measured in kHz for solid state lasers. Higher PRF provides denser point distribution on the surface. Current laser scanners on the market have PRF capabilities of up to or exceeding 150 kHz.
• Beam divergence (mrad): This mode regulates the width (divergence) of the laser beam. It is measured with milliradian (mrad). Typical narrow beam measures 0.3 mrad, while beam with larger footprint; may measure up to 2 mrad14. Narrow beam has a stronger manipulation rate, while wide beam covers larger surfaces.
• Roll compensation (degº): This feature corrects for aircraft roll by biasing the scanner. Scan frequency is sampled and corrections are applied with specific rates (e.g. 5 Hz). Corrections are based on the full scan half-angle of the system. For example, if max half-angle scanning capacity is 30, and 20 is configured, this feature will compensate to a max of 10 at full swing.
• Eye safe altitude (metres): The laser wavelength (λ - nm) and peak power (W) relate to eye safety considerations. Beam divergence (mrad); system PRF (kHz) and flight altitude would also have an influence on eye safety settings.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Table 2 configuration parameters affect LiDAR point density directly. These parameters need to be adjusted to produce the desired final product before a mission. Table 3 illustrates basic computation of cross track, down track and average resolution with point density as a final product. Simple formulas used in the Excel worksheet are explained below.
Table 2. Configuration Parameters for Point Density
Parametre Typical min & max Surface Point Density
Scan Frequency (Hz) 0 – 100 Hz T faster, Thigher
Scan Half-angle (deg) 0 – 30 º T wider, 1 lower
System PRF (kHz) 5 – 150 kHz T faster, Thigher
Operating Altitude (metre) 200 – 4000 m, T higher, 1 lower
Aircraft speed (knots) 10 – 140 knots T faster, 1 lower
• Swath width (m) 4 2* [Altitude * (TAN (half angle) * PI() / 180)]
• Cross track (m) 4 (2 * scan Freq * swath) / system PRF
• Down track (m) 4 (speed / scan freq) / 2
• Resolution (m) 4 (cross track * down track)
• Point density (1/x) 4 1/ cross track * down track
Table 3. Excel computation for LiDAR point density
Requirements LiDAR System Settings
(desired) (desired) Results (calculated)
Altitude (mt) 1700 system PRF (Hz) 70000 Swath (mt) 1237.50
Altitude (feet) 5576 scan Freq (Hz) 50 Crosstrack (mt) 1.77
Speed (knot/ h) 130 scan half angle (±deg) 20 Downtrack (mt) 0.67
Speed (mt/sec) 67 Resolution (mt) 1.09
Overlap (%) 25 Point density (1/ mt) 0.85
Overlap (mt) 309
QA DATA PROCESSING PROCEDURES
LiDAR data processing is a complicated task. Data is large in size and needs powerful hardware and software integration. As to comply with QA procedures, issues need to be addressed and examined for best possible solution prior to any processing practices. Chart 3 illustrates a typical flow of activities for LiDAR data processing.
ASPRS 2009 Annual Conference
Baltimore, Maryland 0 March 9-13, 2009
Figure 3. Data Processing Flow.
Data Accuracy Verification
LiDAR data accuracy is influenced by various sources. Calibration of laser scanner instrument, position and orientation determination system (GPS and IMU) and the alignment between these two subsystems affect the data accuracy significantly. Possible human errors in data handling, transformation and processing may also influence final data accuracy.
Table 4 illustrates accuracy parameters for a popular airborne laser scanner as they are published at manufacturer’s Web site15.
American Society for Photogrammetry and Remote Sensing (ASPRS) accuracy standards for Spatial Data are measured with RMSE statistics16. Accuracy is reported in ground distances at 95% confidence level. DEM created from LiDAR derived data should have a maximum RMSE of 15 cm, which is roughly equivalent to 30 cm accuracy. Therefore, 15 cm RMSE is often referred to as ‘30-cm accuracy at 95-Percent confidence level’. According to Federal Emergency Management Agency (FEMA) 16, the RMSE calculated from a sample of test points will not be the RMSE of the DEM. In fact, for each major classification category, a sample of points should be evaluated independently. These points shall be selected carefully from areas of highest PDOP values registered during survey.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Handling and Management Issues
LiDAR data is large due to its nature. This is considered a limitation for LiDAR data handling and processing. A large area survey may produce hundreds of gigabytes of data. However, in the last few years, advances in computing and digital storage hardware enabled much faster and more efficient handling of
Table 5. Comparison of LiDAR data format sizes
Format/ Resolution 1-mt 2-mt 3-mt 4-mt 5-mt
ASCII 30 MB 7.5 MB 3 MB 2 MB 1 MB
Binary
(LAS) 4 MB 1 MB 0.4 MB 0.25 MB 0.15 MB
LiDAR data. Most LiDAR projects make use of large capacity and high speed external hard disks for data handling. Also, accepting LAS (log ASCII) format as an industry standard for data exchange reduced the size of data significantly. 1 sq km of processed data17 size is compared with Table 5.
Filtering, Segmentation and Classification Process
Ground elevation data is required in most projects. In order to generate bare earth, all elevated features (e.g. vegetation, buildings, bridges and other structures) should be removed from LiDAR data. Segmentation of laser points is to group the neighbouring laser points that have common characteristics18. Elimination of all elevated features above ground may be considered as classification. Main categories of classification are; bare earth and low grass, high grass and crops, brush lands and low trees, forests and urban areas. Various other main or sub-categories may be created depending on the project requirements.
Process, Manipulate and Visualize Data
Currently, vast majority of raw data processing and manipulation is carried out by using proprietary software developed independently by researchers, data providers or sensor manufacturers19. Such software is not available as a separate package causing a limited understanding and manipulation process of raw data by end users. However, after raw data processing is completed, LiDAR data may be manipulated in various ways due to its open format nature. Today, there are several software packages on the market with various tools and functionalities (e.g. TerraSolid/Bentley, Z-I Imaging, ESRI ArcGIS, PCI Geomatics, ER Mapper, QT Modeler, etc). Selection of such software depends on the user needs, budgeting and previous expertise.
PROJECT DELIVERABLES
LiDAR System Report
Final system delivery and calibration package should include reports such as data acquisition and processing methods, pre-mission flight plans, final system configuration parameters, accuracy of LiDAR and control data, discovery and treatment of blunders and other supporting documentation.
Errors report: Blunders, systematic and random errors need to be properly examined and documented. A blunder is an error of major proportion, normally identified and removed during editing or QC process. Systematic errors follow some fixed pattern and are introduced by data collection procedures or system errors. They are predictable in theory, and therefore not like random errors where they follow a normal distribution.
95th Percentile Error: For supplemental and consolidated accuracy tests, this method shall be employed to determine accuracy. The 95th percentile indicates that 95 percent of errors in the dataset will have absolute values of equal or lesser value while 5 percent of values will be of larger value20.
A thorough analysis of errors in the data should be provided with the final report. Any large errors evident in the data should not be deleted before any relevant investigations for the cause is addressed. Every major and minor error found and investigated in the dataset should be documented for inclusion in the QA model.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Flight Report
A detailed flight log should be prepared for every flight mission by LiDAR Operator. Any information relevant to data collection with start/stop time, configuration parameters, flight altitude, heading, PDOP and flight plan used should be documented. This information is necessary
to understand the
characteristics of laser strips during data processing and QC process. A typical flight log is illustrated with Table 6.
Ground Control Report
All information related to ground control network, every control points used and GPS base station monument information should be included in the report. Additional
information such as
local/provincial network that is
Table 6. Flight Log Information
Date 31 January 2010 Julian Date 31
Project Number 711-2010-3 Mission 3rd segment – 2nd try
Aircraft C-GAWS Operator name Jack S.
Pilot name Steve R. Co-Pilot Name David T.
Airport Up CYYJ Airport Down CYYJ
Wheels Up 14:50 Wheels Down 17:10
Aircraft HOBBS start 2345 Aircrafts HOBBS end 2347
Ground
Temperature +15 C Visibility Clear
Cloud cover 12,000 ft – broken Precipitation None
Wind direction 250º - SW Wind speed 22 knots
Atmospheric Pressure 29.96 mbar Other notes
used to tie control points, GPS base station monument metadata (e.g. survey date, accuracy, Datum) and any other facts considered to assist post-mission activities should supplement the report.
Calibration Report
Contractor should provide an initial calibration report that is delivered with system acceptance. Also, contractor must submit evidence and final calibration parameters prior to project initiation for the purposes of identifying and correcting systematic biases21. Initial and final calibration parameters comparison, all adjustments and iteration parameters used during calibration should be made available in the final report.
Delivery of LiDAR Data
All raw datasets and processed LiDAR data (ASCII or LAS) should be supplied. Each processed point should have GPS time, orthometric height, intensity value and positional information (x, y, and z). Break lines must be produced and break line information must be identified with proper flagging and information relevant to their source and accuracy.
DEM data should be delivered with USGS DEM or ASCII XYZ format unless project requires another specific format. The contractor must submit raw datasets in tiles or as a separate file for each data strip. Due to the current limitations with LiDAR data processing software and hardware, each file size is not advised to exceed 1 gigabyte.
All project deliverables must conform to pre-mission specified contract requirements. Datum, projection and coordinate system employed must be documented. Data may be delivered in various forms according to project requirements. Since most organizations have secure and fast broadband connections, online delivery of large amounts of data is becoming more popular than ever. Data delivery with an external hard drive is also commonly used and accepted. If project deliverables require data copied on media (CD or DVD), this practice should conform to ISO 6990 standards.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
CONCLUSION AND RECOMMENDATIONS
Quality Assurance goes a long way with LiDAR mission planning. As noted throughout this paper, this is rather a systematic approach with guidelines than a practical experience with no documentation. There are basic fundamentals as practised daily and in variable nature, and there are project requirements that are addressed very specifically. Some QA procedures are practiced with former expertise of personnel, with no established documentation and standards. As expected, many required practices are overlooked or even forgotten with the urgency of mission completion, leading to project deadline delays, additional expenses and yet mission failures.
This paper presents essential guidelines of QA practices for LiDAR mission planning as the author recommends. Practising any of these guidelines is up to sole decision of the contractor, but highly advised for possible successful mission accomplishment. Author believes that recommendations presented throughout this paper are in general perspective and may be customized with minor modifications to fit specific project requirements. Since many aspects of QA procedures need each other for conclusion, it is necessary not to disregard any steps. Due to the nature or airborne LiDAR surveys, procedures that may appear to be insignificant or unnecessary could result in disappointing final results.
As a final statement, author believes that it is a good practice to remind the LiDAR community with the importance of QA procedures to comply with “Do it right, the first time’.
REFERENCES
1 Habib A.F. and R.W.T. Cheng, 2006. Surface matching strategy for quality control of LiDAR data, Innovations in 3D Geo Information Systems, Lecture notes on Geo Information and Cartography, pp 67-83, Editors: Abdul-Rahman, A., Zlatonova, S., Coors, V., Springer.
2 Csanyi, N. and C.K. Tooth, 2007. Improvement of LiDAR data accuracy using LiDAR-specific ground targets, Photogrammetric Engineering & Remote Sensing, Vol. 73, No. 4, April 2007, pp. 385-396.
3 Base Mapping and Geomatic Services, 2006. LiDAR Specifications, Ministry of Agriculture and Lands, Integrated Land Management Bureau, Ver. 0.5, November 2006, pp. 25-30.
4 Toth C.K., D.A. Grejner-Brzezinska and M. Bevis, 2006. High resolution airborne LiDAR/CC mapping of San Andreas fault, Conference Proceeding, 3rd IAG, 12th FIG Symposium, Baden, Germany.
5 Hodgson, M.E. and P. Bresnahan, 2004. Accuracy of airborne LiDAR-derived elevation: Empirical assessment and error budget, Photogrammetric Engineering & Remote Sensing, Vol. 70, No. 3, pp. 331-339.
6 Maas, H.G., 2003. Planimetric and height accuracy of airborne laser scanner data: User requirements and system performance, Proceedings 49, Photogrammetric Week (Ed. D. Fritsch), Wichmann Verlag, pp. 117-125.
7 SEM WSEM36, ARIND, 2008. Dedication Beyond Expectation, GPS Satellite Almanacs and SEM Program, December 12, 2008. http://www.arinc.com/gps/gpsapps/sem.html.
8 Canadian Spatial Reference System, 2008. Natural Resources Canada, 3 November 2008. http://www.geod.nrcan.gc.ca/index e.php.
9 Products, 2008. BCACS GPS Base Sation Data, Crown Registry and Geographic Base, GeoBC, December 12, 2008. http://www.ilmb.gov.bc.ca/bmgs/products/geospatial/bcacs.htm.
10 SNB, 2007. New Brunswick Control Network, Terms and Conditions of Use, 2007. https://www.pxw1.snb.ca/snb7001/e/2000/2923e.asp.
11 CORS, 2008. Continuously Operating Reference Stations, National Geodetic Survey, National and Cooperative CORS, http://www.ngs.noaa.gov/CORS.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
12 Hutton, J, T. et al., 2007. New Developments of Inertial Navigation Systems at Applanix, Online Report Applanix Corporation, Markham, ON, http://www.applanix.com/media/downloads/products/articles_papers.
13 GrafNav, 2007. Aerial Survey Base, GrafNav Online Report ver 7.8x, August, USA, http://www.aerial-survey-base.com/PDF/PPP Processing.pdf.
14 Wehr, A. and U. Lohr, 1999. Airborne laser scanning – An introduction and overview, ISPRS Journal of Photogrammetry and Remote Sensing, Vol.54, No. 2-3, pages 68-82.
15 Optech, 2008. ALTM 3100EA – Enhanced Accuracy, Online Presentation, Vaughan, ON, Canada, http://www.optech.ca/pdf/Brochures/ALTM3100EAwspecsfnl.pdf.
16 FEMA, 1998.Geospatial Positioning Accuracy Standards Part 3: National Standard for Spatial Data Accuracy, FGDC-STD-007.3-1998, Virginia.
17 NOAA, 2008. Remote Sensing for Coastal Management, National Ocean Service, online document, Charleston, SC, USA, http://www.csc.noaa.gov/crs/rs_apps/sensors/lidar.htm#data_acquisition.
18 Morgan, M and A.F. Habib, 2002. Interpolation of LiDAR data and automatics building extraction, ACSM-ASPRS Annual Conference Proceedings, Washington, DC.
19 Flood, M., 2001. LiDAR activities and research priorities in the commercial sector, International Archives of Photogrammetry and Remote Sensing, Volume XXXIV-3/W4 Annapolis, MD, 22-24 Oct.
20 ASPRS LiDAR Committee, 2004. ASPRS Guidelines: Vertical Accuracy Reporting for LiDAR Data, Flood, M. (ed), Online report, May, Maryland, USA.
21 Federal Geographic Data Committee, 2000. Airborne Light Detection and Ranging (LIDAR) Systems, Appendix
4B, Online Report: Department of Homeland Security, May, US,
http://www.fema.gov/library/viewRecord.do?id=2345.
ASPRS 2009 Annual Conference
Baltimore, Maryland March 9-13, 2009
Effective Date: 2012/06/01
Number: PD- 35
Title:
Tidak ada komentar:
Posting Komentar