Unsupervised Domain Adaptation Using Regularized Hyper-Graph Matching

Abstract

Domain adaptation (DA) addresses the real-world image classification problem\nof discrepancy between training (source) and testing (target) data\ndistributions. We propose an unsupervised DA method that considers the presence\nof only unlabelled data in the target domain. Our approach centers on finding\nmatches between samples of the source and target domains. The matches are\nobtained by treating the source and target domains as hyper-graphs and carrying\nout a class-regularized hyper-graph matching using first-, second- and\nthird-order similarities between the graphs. We have also developed a\ncomputationally efficient algorithm by initially selecting a subset of the\nsamples to construct a graph and then developing a customized optimization\nroutine for graph-matching based on Conditional Gradient and Alternating\nDirection Multiplier Method. This allows the proposed method to be used widely.\nWe also performed a set of experiments on standard object recognition datasets\nto validate the effectiveness of our framework over state-of-the-art\napproaches.\n