On Localized Discrepancy for Domain Adaptation

Abstract

We propose the discrepancy-based generalization theories for unsupervised domain adaptation. Previous theories introduced distribution discrepancies defined as the supremum over complete hypothesis space. The hypothesis space may contain hypotheses that lead to unnecessary overestimation of the risk bound. This paper studies the localized discrepancies defined on the hypothesis space after localization. First, we show that these discrepancies have desirable properties. They could be significantly smaller than the pervious discrepancies. Their values will be different if we exchange the two domains, thus can reveal asymmetric transfer difficulties. Next, we derive improved generalization bounds with these discrepancies. We show that the discrepancies could influence the rate of the sample complexity. Finally, we further extend the localized discrepancies for achieving super transfer and derive generalization bounds that could be even more sample-efficient on source domain.

References in corpus (11)

• Learning Transferable Features with Deep Adaptation Networks
• Conditional Adversarial Domain Adaptation
• Domain Adaptation: Learning Bounds and Algorithms
• A PAC-Bayesian bound for Lifelong Learning
• The Benefit of Multitask Representation Learning
• Understanding and Improving Information Transfer in Multi-Task Learning
• Support and Invertibility in Domain-Invariant Representations
• Hypothesis Transfer Learning via Transformation Functions
• Provable Guarantees for Gradient-Based Meta-Learning
• A New PAC-Bayesian Perspective on Domain Adaptation
• On the Value of Target Data in Transfer Learning

Cited by in corpus (2)

• Adaptive Personalized Federated Learning
• Discrepancy-Based Active Learning for Domain Adaptation