Robust visual domain adaptation with low-rank reconstruction

摘要: Visual domain adaptation addresses the problem of adapting sample distribution source to target domain, where recognition task is intended but data distributions are different. In this paper, we present a low-rank reconstruction method reduce disparity. Specifically, transform visual samples in into an intermediate representation such that each transformed can be linearly reconstructed by domain. Unlike existing work, our captures intrinsic relatedness during process while uncovering noises and outliers cannot adapted, making it more robust than previous methods. We formulate as constrained nuclear norm l 2, 1 minimization objective then adopt Augmented Lagrange Multiplier (ALM) for optimization. Extensive experiments on various tasks show proposed consistently significantly beats state-of-the-art