A convex optimization approach to semi-supervised identification of switched ARX systems

Abstract

This paper proposes a general convex framework for robustly identifying discrete-time affine hybrid systems from measurements contaminated by noise (both process and measurement) and outliers. Our main result shows that this problem can be formulated as a constrained polynomial optimization, for which a monotonically convergent sequence of tractable convex relaxations can be obtained by exploiting recent developments in sparse polynomial optimization. A salient feature of the proposed framework is its ability to incorporate existing a-priori information about the noise, co-ocurrences, or the switching sequence. These results are illustrated with several examples showing the ability of the proposed approach to make effective use of this additional information. 53rd IEEE Conference on Decision and Control

Chordal Decomposition in Rank Minimized Semidefinite Programs with Applications to Subspace Clustering

“…that assign each point x j to the subspace spanned by the normal vector r i . This problem arises in many practical applications including switched system identification from noisy input/output data (see e.g., [24], [25]), where each subspace normal is defined by the coefficients of each switching system. In the general formulation, a point x j belongs to subspace i if r T i x j = 0.…”

Compressive Information Extraction: A Dynamical Systems Approach1

“…where all the matrices involved are affine in the optimization variables m. At this point, a tractable convex relaxation can be obtained by using the nuclear norm as a surrogate for rank, leading to a convex semidefinite program that can be solved using widely available tools (see [24,32] for details).…”

Classification of dynamics on Riemannian manifold

“…In this case, the problem is known to be NP hard and most existing methods are based upon convex relaxations of the original non-convex problem. In particular, [21,109,110] proposed a polynomial optimization based approach, which in turn is relaxed to a sequence of convex semi-definite programs. While in principle this sequence of approximations is guaranteed to converge to the actual system, in practice, computational complexity prevents considering higher elements of the sequence, necessitating adding rank constraints to the algorithm.…”

“…These results are illustrated with an academic example and a non-trivial practical one: activity segmentation from time traces of the position of a person's centroid. While the proposed technique is less general than those in [21,109,110], since it assumes dwell time constraints, it is able to exploit these constraints to substantially lower the computational burden, as opposed to polynomial optimization based approaches. Thus, it substantially outperforms these approaches in many practical scenarios (e.g.…”