Domain Adaptation with Asymmetrically-Relaxed Distribution Alignment

Abstract

Domain adaptation addresses the common problem when the target distribution generating our test data drifts from the source (training) distribution. While absent assumptions, domain adaptation is impossible, strict conditions, e.g. covariate or label shift, enable principled algorithms. Recently-proposed domain-adversarial approaches consist of aligning source and target encodings, often motivating this approach as minimizing two (of three) terms in a theoretical bound on target error. Unfortunately, this minimization can cause arbitrary increases in the third term, e.g. they can break down under shifting label distributions. We propose asymmetrically-relaxed distribution alignment, a new approach that overcomes some limitations of standard domain-adversarial algorithms. Moreover, we characterize precise assumptions under which our algorithm is theoretically principled and demonstrate empirical benefits on both synthetic and real datasets.

References in corpus (6)

• Deep Domain Confusion: Maximizing for Domain Invariance
• Domain Separation Networks
• Domain Adaptation: Learning Bounds and Algorithms
• Wasserstein Distance Guided Representation Learning for Domain Adaptation
• Detecting and Correcting for Label Shift with Black Box Predictors
• Analysis of Kernel Mean Matching under Covariate Shift

Cited by in corpus (19)

• Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems
• Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction
• WILDS: A Benchmark of in-the-Wild Distribution Shifts
• Learning Robust Global Representations by Penalizing Local Predictive Power
• Rethinking Distributional Matching Based Domain Adaptation
• Domain Adaptation for One-Class Classification: Monitoring the Health of Critical Systems Under Limited Information
• Robust Local Preserving and Global Aligning Network for Adversarial Domain Adaptation
• Cluster Alignment with a Teacher for Unsupervised Domain Adaptation
• Minimax Lower Bounds for Transfer Learning with Linear and One-hidden Layer Neural Networks
• Modeling Multiple Views via Implicitly Preserving Global Consistency and Local Complementarity
• Learning Invariant Representations and Risks for Semi-supervised Domain Adaptation
• A Theory of Label Propagation for Subpopulation Shift
• An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation
• Class-imbalanced Domain Adaptation: An Empirical Odyssey
• Barycentric-alignment and reconstruction loss minimization for domain generalization
• Tracking the risk of a deployed model and detecting harmful distribution shifts
• A Dictionary Approach to Domain-Invariant Learning in Deep Networks
• Information Theory-Guided Heuristic Progressive Multi-View Coding
• The Role of Embedding Complexity in Domain-invariant Representations