PROJECT TITLE :

Multi-View Intact Space Learning

ABSTRACT:

It's sensible to assume that an individual view is unlikely to be sufficient for effective multi-view learning. Therefore, integration of multi-read data is each valuable and necessary. During this paper, we tend to propose the Multi-view Intact Area Learning (MISL) algorithm, that integrates the encoded complementary information in multiple views to find a latent intact representation of the data. Even though every read on its own is insufficient, we have a tendency to show theoretically that by combing multiple views we have a tendency to can acquire abundant information for latent intact space learning. Using the Cauchy loss (a way employed in statistical learning) because the error measurement strengthens robustness to outliers. We propose a brand new definition of multi-read stability and then derive the generalization error bound based mostly on multi-read stability and Rademacher complexity, and show that the complementarity between multiple views is beneficial for the stability and generalization. MISL is efficiently optimized employing a novel Iteratively Reweight Residuals (IRR) technique, whose convergence is theoretically analyzed. Experiments on synthetic knowledge and real-world datasets demonstrate that MISL is a good and promising algorithm for practical applications.