Self-Challenging Improves Cross-Domain Generalization

Abstract

Convolutional Neural Networks (CNN) conduct image classification by activating dominant features that correlated with labels. When the training and testing data are under similar distributions, their dominant features are similar, which usually facilitates decent performance on the testing data. The performance is nonetheless unmet when tested on samples from different distributions, leading to the challenges in cross-domain image classification. We introduce a simple training heuristic, Representation Self-Challenging (RSC), that significantly improves the generalization of CNN to the out-of-domain data. RSC iteratively challenges (discards) the dominant features activated on the training data, and forces the network to activate remaining features that correlates with labels. This process appears to activate feature representations applicable to out-of-domain data without prior knowledge of new domain and without learning extra network parameters. We present theoretical properties and conditions of RSC for improving cross-domain generalization. The experiments endorse the simple, effective and architecture-agnostic nature of our RSC method.

to appear at ECCV2020 as an oral paper

References in corpus (5)

• Deep Visual Domain Adaptation: A Survey
• Domain Generalization via Invariant Feature Representation
• DropBlock: A regularization method for convolutional networks
• Generalizing to Unseen Domains via Adversarial Data Augmentation
• Learning Robust Representations by Projecting Superficial Statistics Out

Cited by in corpus (25)

• Towards Out-Of-Distribution Generalization: A Survey
• Comprehensive Attention Self-Distillation for Weakly-Supervised Object Detection
• Learning Debiased Representation via Disentangled Feature Augmentation
• FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space
• Federated Learning with Domain Generalization
• Test-time Batch Statistics Calibration for Covariate Shift
• SAND-mask: An Enhanced Gradient Masking Strategy for the Discovery of Invariances in Domain Generalization
• Gradient Matching for Domain Generalization
• A Fourier-based Framework for Domain Generalization
• Discrete Representations Strengthen Vision Transformer Robustness
• Open Domain Generalization with Domain-Augmented Meta-Learning
• Model-Based Domain Generalization
• Learning to Diversify for Single Domain Generalization
• Calibrated Feature Decomposition for Generalizable Person Re-Identification
• Domain Generalization via Inference-time Label-Preserving Target Projections
• Causality-inspired Single-source Domain Generalization for Medical Image Segmentation
• Cooperative Training and Latent Space Data Augmentation for Robust Medical Image Segmentation
• Unsupervised Domain Generalization for Person Re-identification: A Domain-specific Adaptive Framework
• On the Challenges of Open World Recognitionunder Shifting Visual Domains
• Explainability-aided Domain Generalization for Image Classification
• Confounder Identification-free Causal Visual Feature Learning
• Encouraging Intra-Class Diversity Through a Reverse Contrastive Loss for Better Single-Source Domain Generalization
• Reappraising Domain Generalization in Neural Networks
• Domain-Specific Bias Filtering for Single Labeled Domain Generalization
• Semi-Supervised Domain Generalization with Evolving Intermediate Domain