Translator Disclaimer
14 May 2018 Selective erasures for high-dimensional robust subspace tracking
Author Affiliations +
Abstract
This paper presents an online method to track a subspace U from severely corrupted and incomplete data. If we could identify the corrupted entries in a new observation x, then we would be able to update U according to the uncorrupted entries in x using an incomplete-data rank-one update. The challenge is to identify the corrupted entries in x, which is in general NP-hard. To work around this we propose an approach that iteratively removes the entries that most affect partial projections of x onto U. Our experiments show that this simple approach outperforms state-of-the-art methods, including ℓ1-optimization, specially when most entries in x are corrupted.
© (2018) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Daniel L. Pimentel-Alarcon "Selective erasures for high-dimensional robust subspace tracking", Proc. SPIE 10658, Compressive Sensing VII: From Diverse Modalities to Big Data Analytics, 1065808 (14 May 2018); https://doi.org/10.1117/12.2311891
PROCEEDINGS
11 PAGES


SHARE
Advertisement
Advertisement
RELATED CONTENT


Back to Top