Deep Domain Adaptation by Joint Multi-order Domain Distribution Difference

Deep learning has been applied to transfer learning. This deep transfer learning is based on learning transferable deep adaptive networks as a framework to reduce distribution differences between domains. Existing domain adaptive methods cannot use simple domain distribution matching information, so they cannot always compensate the algorithm performance degradation caused by domain offset. In this paper, we propose a joint multi-order domain distribution difference method (JDD), which uses Gaussian kernel functions to jointly unify the multi-order saturation domain distribution difference information, thereby learning more domain-invariant features end-to-end and solving the problem more significantly cross-domain alignment issues. The JDD method can be applied to actual migration tasks and is easy to train and implement. Experiments show that our method has better classification accuracy than existing adaptive methods.