Maximum Density Divergence for Domain Adaptation

Unsupervised domain adaptation addresses the problem of transferring knowledge from a well-labeled source to an unlabeled target where two domains have distinctive data distributions. Thus, essence is mitigate distribution divergence between domains. The state-of-the-art methods practice this very idea by either conducting adversarial training or minimizing metric which defines gaps. In paper, we propose new method named Adversarial Tight Match (ATM) enjoys benefits both and learning. Specifically, at first, novel distance loss, Maximum Density Divergence (MDD), quantify divergence. MDD minimizes inter-domain ("match" in ATM) maximizes intra-class density ("tight" ATM). Then, address equilibrium challenge issue adaptation, consider leveraging proposed into framework. At last, tailor as practical learning loss report our ATM. Both empirical evaluation theoretical analysis are reported verify effectiveness method. experimental results on four benchmarks, classical large-scale, show that able achieve performance most evaluations. Codes datasets used paper available {\it github.com/lijin118/ATM}.