Title :
Sample complexity for topology estimation in networks of LTI systems
Author :
Tan, Vincent Y F ; Willsky, Alan S.
Author_Institution :
Dept. of Electr. & Comput. Eng., Univ. of Wisconsin-Madison, Madison, WI, USA
Abstract :
This paper proposes a consistent and computationally efficient FFT-based algorithm for inferring the network topology where each node in the network is associated to a wide-sense stationary, ergodic, Gaussian process. Each edge of the tree network is characterized by a linear, time-invariant dynamical system and additive white Gaussian noise. The proposed algorithm uses Bartlett´s procedure to produce periodogram estimates of cross power spectral densities between processes. Under appropriate assumptions, we prove that the number of vector-valued samples from a single sample path required for consistent estimation is polylogarithmic in the number of nodes in the network. Thus, the sample complexity is low. Our proof uses properties of spectral estimates and analysis for learning tree-structured graphical models.
Keywords :
AWGN; Gaussian processes; computational complexity; fast Fourier transforms; learning (artificial intelligence); trees (mathematics); Bartlett procedure; FFT-based algorithm; LTI network system; additive white Gaussian noise; cross power spectral densities; learning tree-structured graphical models; linear time-invariant dynamical system; network topology; periodogram estimation; polylogarithmic; sample complexity; topology estimation; tree network edge; vector-valued samples; wide-sense stationary ergodic Gaussian process; Approximation algorithms; Complexity theory; Entropy; Estimation; Network topology; Stochastic processes; Topology;
Conference_Titel :
Decision and Control and European Control Conference (CDC-ECC), 2011 50th IEEE Conference on
Conference_Location :
Orlando, FL
Print_ISBN :
978-1-61284-800-6
Electronic_ISBN :
0743-1546
DOI :
10.1109/CDC.2011.6160412