A novel domain adaptation theory with Jensen–Shannon divergence

摘要

Domain adaptation aims to alleviate the shift between training and test distribution, where the DA theory is crucial in understanding the success of domain adaptation algorithms. In this paper, we reveal the incoherence between the empirical domain adversarial training and its generally assumed theoretical counterpart based on H-divergence. Concretely, we find that H-divergence is not equivalent to Jensen–Shannon divergence, the optimization objective in domain adversarial training. To this end, we establish a new theoretical framework by directly proving the upper and lower target risk bounds based on the joint distributional Jensen–Shannon divergence. We further derive bidirectional upper bounds for marginal and conditional shifts. Our framework exhibits inherent flexibility for different transfer learning problems, which is usable for various scenarios. From an algorithmic perspective, our theory enables a generic guideline of the unified principles of semantic conditional matching, feature marginal matching, and label marginal shift correction. We employ algorithms for each principle and empirically validate the benefits of our framework.