Pursuit of Low-Rank Models of Time-Varying Matrices Robust to Sparse and Measurement Noise

TL;DR

This paper introduces a scalable method for tracking low-rank models of time-varying matrices that is robust to both measurement noise and sparse noise, with theoretical error bounds and practical validation on a benchmark dataset.

Contribution

It proposes a novel robust tracking method using randomized coordinate descent for low-rank matrix models, with theoretical error bounds and empirical validation.

Findings

Effective robustness to measurement and sparse noise

Scalable approach suitable for large datasets

Successful application on change detection benchmark

Abstract

In tracking of time-varying low-rank models of time-varying matrices, we present a method robust to both uniformly-distributed measurement noise and arbitrarily-distributed ``sparse'' noise. In theory, we bound the tracking error. In practice, our use of randomised coordinate descent is scalable and allows for encouraging results on changedetection net, a benchmark.