Multiple View Locality Preserving Projections with Pairwise Constraints

Abstract

In this paper, we consider a general problem of multiple view learning from pairwise constraints in the form of must-link and cannot-link. As one kind of side information, the must-link constraints imply that a pair of instances belongs to the same class, while the cannot-link constraints compel them to be different classes. Given must-link and cannot-link constraints, we propose a new locality preserving projections approach for multiple view semi-supervised dimensionality reduction. The goal of the proposed approach uses pairwise constraints to derive embedding in each view together with the unlabeled instances. Hence, the consensus pattern can be learned from multiple embeddings of multiple representations. Experimental results on real-world data sets show the effectiveness of the proposed algorithm.