Orthogonal Multi-View Subspace Learning

Abstract: Multi-view data are increasingly collected for a variety of applications in the real world. They contain complementary, redundant, and corroborative contents and so provide more informative than single_x005fview data when it comes to characterize objects of the real-world. It is rather natural for human beings to perceive the world through comprehensive information collected by multiple sensory organs, but it is an open question on how to endow machines with analogous cognitive capabilities to do the same.One of the fundamental challenges is how to represent and summarize multi-view data in such a way that comprehensive information concealed in multi-view data can be properly exploited by multi-view learning models. In this talk, we will present a unified framework for multi-view subspace learning to learn individual orthogonal projections for all views. The framework integrates the correlations within multiple views, supervised discriminant capacity, and distance preservation in a concise and compact way. It not only includes several existing models as special cases, but also inspires new novel models. Besides the framework, we will discuss highly efficient numerical methods to solve the associated optimization problems. The methods are built upon an iterative Krylov subspace method which can easily scale up for large size datasets. Extensive experiments are conducted on various real-world datasets for the multi-view discriminant analysis and multi-view multi-label classification tasks. The experimental results demonstrate that the proposed models are consistently competitive to and often better than the state-of-the-art methods.This is a joint work with Li Wang (UT Arlington), Lei-hong Zhang (Soochow University), and Chungen Shen (University of Shanghai for Science and Technology).

Mathematics

Audience: researchers in the topic

ICCM 2020