On Learning Invariant Representation for Domain Adaptation

Abstract

Due to the ability of deep neural nets to learn rich representations, recent advances in unsupervised domain adaptation have focused on learning domain-invariant features that achieve a small error on the source domain. The hope is that the learnt representation, together with the hypothesis learnt from the source domain, can generalize to the target domain. In this paper, we first construct a simple counterexample showing that, contrary to common belief, the above conditions are not sufficient to guarantee successful domain adaptation. In particular, the counterexample exhibits \emph{conditional shift}: the class-conditional distributions of input features change between source and target domains. To give a sufficient condition for domain adaptation, we propose a natural and interpretable generalization upper bound that explicitly takes into account the aforementioned shift. Moreover, we shed new light on the problem by proving an information-theoretic lower bound on the joint error of \emph{any} domain adaptation method that attempts to learn invariant representations. Our result characterizes a fundamental tradeoff between learning invariant representations and achieving small joint error on both domains when the marginal label distributions differ from source to target. Finally, we conduct experiments on real-world datasets that corroborate our theoretical findings. We believe these insights are helpful in guiding the future design of domain adaptation and representation learning algorithms.

Compared with the last version, the current one adds a new corollary for the case of different feature transformations (encoders) on source/target domains. Fix a typo in Fig. 1

References in corpus (8)

• Learning Transferable Features with Deep Adaptation Networks
• FCNs in the Wild: Pixel-level Adversarial and Constraint-based Adaptation
• CyCADA: Cycle-Consistent Adversarial Domain Adaptation
• Domain Separation Networks
• Adversarial Discriminative Domain Adaptation
• Domain-Adversarial Neural Networks
• Multiple Source Adaptation and the Renyi Divergence
• Support and Invertibility in Domain-Invariant Representations

Cited by in corpus (40)

• SCL: Towards Accurate Domain Adaptive Object Detection via Gradient Detach Based Stacked Complementary Losses
• In Search of Lost Domain Generalization
• Support and Invertibility in Domain-Invariant Representations
• Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation
• Rethinking Distributional Matching Based Domain Adaptation
• Domain Aggregation Networks for Multi-Source Domain Adaptation
• Robust Local Preserving and Global Aligning Network for Adversarial Domain Adaptation
• Understanding Self-Training for Gradual Domain Adaptation
• Vicinal and categorical domain adaptation
• Semi-supervised Models are Strong Unsupervised Domain Adaptation Learners
• Learning Overlapping Representations for the Estimation of Individualized Treatment Effects
• Minimax Lower Bounds for Transfer Learning with Linear and One-hidden Layer Neural Networks
• Flow imaging as an alternative to pressure transducers through vision transformers and convolutional neural networks
• Anomaly Detection with Domain Adaptation
• Improving Unsupervised Domain Adaptation with Variational Information Bottleneck
• Curriculum Manager for Source Selection in Multi-Source Domain Adaptation
• MTS-CycleGAN: An Adversarial-based Deep Mapping Learning Network for Multivariate Time Series Domain Adaptation Applied to the Ironmaking Industry
• Learning to Match Distributions for Domain Adaptation
• Model-based Policy Optimization with Unsupervised Model Adaptation
• Continuous Transfer Learning with Label-informed Distribution Alignment
• Understanding the Limits of Unsupervised Domain Adaptation via Data Poisoning
• Cross-Domain Sentiment Classification with In-Domain Contrastive Learning
• Adaptive Invariance for Molecule Property Prediction
• Unsupervised Domain Adaptation with Progressive Domain Augmentation
• Gradual Domain Adaptation via Self-Training of Auxiliary Models
• Learning Invariant Representation with Consistency and Diversity for Semi-supervised Source Hypothesis Transfer
• An Information-theoretic Approach to Distribution Shifts
• Multi-task Learning by Leveraging the Semantic Information
• Discriminative Active Learning for Domain Adaptation
• Target Consistency for Domain Adaptation: when Robustness meets Transferability
• Multi-Purposing Domain Adaptation Discriminators for Pseudo Labeling Confidence
• Sim2Real for Self-Supervised Monocular Depth and Segmentation
• Interventional Domain Adaptation
• Conditional Bures Metric for Domain Adaptation
• Cross-domain error minimization for unsupervised domain adaptation
• NI-UDA: Graph Adversarial Domain Adaptation from Non-shared-and-Imbalanced Big Data to Small Imbalanced Applications
• Task Guided Compositional Representation Learning for ZDA
• The Role of Embedding Complexity in Domain-invariant Representations
• Domain Adaptive Semantic Segmentation without Source Data
• Towards Uncovering the Intrinsic Data Structures for Unsupervised Domain Adaptation using Structurally Regularized Deep Clustering