Semi-supervised manifold alignment with multi-graph embedding

Abstract

In this paper, a novel semi-supervised manifold alignment approach via multiple graph embeddings (MA-MGE) is proposed. Different from the traditional manifold alignment algorithms that use a single graph embedding to learn the latent manifold structure of each data set, our approach utilizes multiple graph embeddings to learn a joint latent manifold structure. Therefore a composite manifold representation with complete and more useful information is obtained from each dataset through a dynamic reconstruction of multiple graphs. Also, an optimization strategy based on eigen-value solutions is provided. Experimental results on Protein, COIL-20 and Face-10 datasets demonstrate superior performance of the proposed method compared with the state-of-the-art methods, such as semi-supervised manifold alignment (SSMA), manifold alignment using Procrustes analysis (PAMA) and manifold alignment without correspondence (UNMA).