AdaMatch: A Unified Approach to Semi-Supervised Learning and Domain Adaptation

We extend semi-supervised learning to the problem of domain adaptation to significantly higher-accuracy models that train on one data distribution test on a different one. With the goal of generality, we introduce, a method that unifies the tasks of unsupervised domain adaptation (UDA), semi-supervised learning (SSL), and semi-supervised domain adaptation (SSDA). In an extensive experimental study, we compare its behavior with state-of-the-art techniques from SSL, SSDA, and UDA on vision tasks. We find AdaMatch either matches or significantly exceeds state-of-the-art in each case using the same hyper-parameters regardless of dataset or task. For example, AdaMatch nearly doubles the accuracy compared that of the prior state-of-the-art on the UDA task for DomainNet and even the accuracy of the prior state-of-the-art obtained with pre-training 6. 4% when AdaMatch is trained completely from scratch. Furthermore, by AdaMatch with just one labeled example per class from the target (i. e. , the SSDA setting), we increase the target accuracy by an 6. 1%, and with 5 labeled examples, by 13. 6%.