THE ELEVENTH IEEE SENSOR ARRAY AND MULTICHANNEL SIGNAL PROCESSING WORKSHOP (IEEE SAM2020)

349 / 2020-01-12 14:49:00

Online Robust Reduced-Rank Regression viaStochastic Majorization-Minimization Method

Multivariate regression; stochastic optimization; heavy-tails; low-rank; large-scale optimization

IEEE SAM 2020: Full Regular Papers > Non-Gaussian, nonlinear, and non-stationary models

The reduced-rank regression (RRR) model is widely used in many data analytics applications where the response variables are believed to depend on a few linear combinations of the predictor variables, or when such linear combinations are of special interest. In this paper, we will address the RRR estimation problem by considering two common issues: 1) the estimation should be robust to heavy-tailed noise or outliers; 2) the estimation should be amenable to the large-scale data set. To address the robustness, a robust maximum likelihood estimation procedure is adopted. To deal with the large-scale problem setting, a stochastic optimization problem is formulated. To solve this stochastic optimization problem, an algorithm based on the stochastic majorization minimization method is proposed. The efficiency of the proposed algorithm is demonstrated by comparing with the state-of-the-art method via numerical simulations.