Timezone: »

A Pseudo-Bayesian Algorithm for Robust PCA
Tae-Hyun Oh · Yasuyuki Matsushita · In So Kweon · David Wipf

Wed Dec 07 09:00 AM -- 12:30 PM (PST) @ Area 5+6+7+8 #82 #None

Commonly used in many applications, robust PCA represents an algorithmic attempt to reduce the sensitivity of classical PCA to outliers. The basic idea is to learn a decomposition of some data matrix of interest into low rank and sparse components, the latter representing unwanted outliers. Although the resulting problem is typically NP-hard, convex relaxations provide a computationally-expedient alternative with theoretical support. However, in practical regimes performance guarantees break down and a variety of non-convex alternatives, including Bayesian-inspired models, have been proposed to boost estimation quality. Unfortunately though, without additional a priori knowledge none of these methods can significantly expand the critical operational range such that exact principal subspace recovery is possible. Into this mix we propose a novel pseudo-Bayesian algorithm that explicitly compensates for design weaknesses in many existing non-convex approaches leading to state-of-the-art performance with a sound analytical foundation.

Author Information

Tae-Hyun Oh (KAIST)

Tae-Hyun Oh is a postdoctoral associate at MIT CSAIL from Aug, 2017. He received the B.E. degree (The highest ranking) in Computer Engineering from Kwang-Woon University, South Korea in 2010, and the M.S. and Ph.D degrees in Electrical Engineering from KAIST, South Korea in 2012 and 2017, respectively. He was a research intern in the Visual Computing group, Microsoft Research, Beijing and in the Cognitive group, Microsoft Research, Redmond in 2014 and 2016, respectively. He was a recipient of Microsoft Research Asia fellowship, a gold prize of Samsung HumanTech thesis award, twice Qualcomm Innovation awards and top research achievement awards from KAIST. His research interests include robust computer vision and machine learning.

Yasuyuki Matsushita (Osaka University)
In So Kweon (KAIST)
David Wipf (Microsoft Research)

More from the Same Authors