Deep Multi-View Learning via Task-Optimal CCA

  • 2019-07-17 20:06:47
  • Heather D. Couture, Roland Kwitt, J. S. Marron, Melissa Troester, Charles M. Perou, Marc Niethammer
  • 24

Abstract

Canonical Correlation Analysis (CCA) is widely used for multimodal dataanalysis and, more recently, for discriminative tasks such as multi-viewlearning; however, it makes no use of class labels. Recent CCA methods havestarted to address this weakness but are limited in that they do notsimultaneously optimize the CCA projection for discrimination and the CCAprojection itself, or they are linear only. We address these deficiencies bysimultaneously optimizing a CCA-based and a task objective in an end-to-endmanner. Together, these two objectives learn a non-linear CCA projection to ashared latent space that is highly correlated and discriminative. Our methodshows a significant improvement over previous state-of-the-art (including deepsupervised approaches) for cross-view classification, regularization with asecond view, and semi-supervised learning on real data.

 

Quick Read (beta)

loading the full paper ...