Distributed robust subspace tracking

Yannis Kopsinis, Symeon Chouvardas, Sergios Theodoridis

Abstract: In this paper, a distributed, set-theoretic based subspace tracking scheme is presented. In particular, each one of the agents in the network has access to a subset of data, which are not allowed to be shared among them. Moreover, the data vectors lie on a low-rank linear subspace, which is unknown and it might also be time-varying. The agents aim at estimating and tracking the unknown subspace using solely their own data and the tentative subspace estimates of their neighbours. Moreover, some of the the data might be corrupted with outlier noise. Method is evaluated in a synthetic simulation example, where the unknown subspace exhibits abrupt changes.