Self-Challenging Improves Cross-Domain Generalization

2020/07/05 by Zeyi Huang, Huang, Zeyi, Haohan Wang +5 · 53 citations
Computer Science · Mathematics · #Advanced Neural Network Applications #Artificial intelligence #Chromatin structure remodeling (RSC) complex #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Convolutional neural network #Domain (mathematical analysis) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Feature (linguistics) #Feature learning #Generalization #Heuristic #Image (mathematics) #Machine Learning (cs.LG) #Machine learning #Mathematics #Multimodal Machine Learning Applications #Pattern recognition (psychology) #Process (computing) #Representation (politics) #Simple (philosophy) #cs.CV #cs.LG

paper · pdf · doi:10.48550/arxiv.2007.02454

published in arXiv (Cornell University) (Cornell University) · to appear at ECCV2020 as an oral paper

arxiv created 2020/07/05 · openalex publication_date 2020/07/05 · arxiv updated 2020/07/08 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/08

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.

Citations

• Domain Generalization via Invariant Feature Representation
• DropBlock: A regularization method for convolutional networks

Cited by

• Federated Domain Generalization with Latent Space Inversion
• Semantic-aware Random Convolution and Source Matching for Domain Generalization in Medical Image Segmentation
• Harnessing Diffusion-Generated Synthetic Images for Fair Image Classification
• ODP-Bench: Benchmarking Out-of-Distribution Performance Prediction
• SAND-mask: An Enhanced Gradient Masking Strategy for the Discovery of Invariances in Domain Generalization
• FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space
• Connecting Domains and Contrasting Samples: A Ladder for Domain Generalization
• MAT-Agent: Adaptive Multi-Agent Training Optimization
• Learning to Diversify for Single Domain Generalization
• Class-Invariant Test-Time Augmentation for Domain Generalization
• Learning Domain- and Class-Disentangled Prototypes for Domain-Generalized EEG Emotion Recognition
• Towards Out-Of-Distribution Generalization: A Survey
• Reinforcement Learning Driven Generalizable Feature Representation for Cross-User Activity Recognition
• Quantization Meets OOD: Generalizable Quantization-aware Training from a Flatness Perspective
• MorphGen: Morphology-Guided Representation Learning for Robust Single-Domain Generalization in Histopathological Cancer Classification
• Domain Generalization of Pathological Image Segmentation by Patch-Level and WSI-Level Contrastive Learning
• Confounder Identification-free Causal Visual Feature Learning
• A Fourier-based Framework for Domain Generalization
• Learning from Heterogeneous Structural MRI via Collaborative Domain Adaptation for Late-Life Depression Assessment
• Federated Learning with Domain Generalization
• Reappraising Domain Generalization in Neural Networks
• Open Domain Generalization with Domain-Augmented Meta-Learning
• Fully Automated SAM for Single-source Domain Generalization in Medical Image Segmentation
• Simulate, Refocus and Ensemble: An Attention-Refocusing Scheme for Domain Generalization
• Domain Generalization via Pareto Optimal Gradient Matching
• FEDTAIL: Federated Long-Tailed Domain Generalization with Sharpness-Guided Gradient Matching
• GeoDistill: Geometry-Guided Self-Distillation for Weakly Supervised Cross-View Localization
• Adversarial Data Augmentation for Single Domain Generalization via Lyapunov Exponent-Guided Optimization
• Deconfounding Causal Inference through Two-Branch Framework with Early-Forking for Sensor-Based Cross-Domain Activity Recognition
• Bridging Domain Generalization to Multimodal Domain Generalization via Unified Representations
• Prompt Disentanglement via Language Guidance and Representation Alignment for Domain Generalization
• Discrete Representations Strengthen Vision Transformer Robustness
• Learning Time-Aware Causal Representation for Model Generalization in Evolving Domains
• Domain-Adversarial Anatomical Graph Networks for Cross-User Human Activity Recognition
• Unsupervised Domain Generalization for Person Re-identification: A Domain-specific Adaptive Framework
• Pseudo Multi-Source Domain Generalization: Bridging the Gap Between Single and Multi-Source Domain Generalization
• Single Domain Generalization for Alzheimer's Detection from 3D MRIs with Pseudo-Morphological Augmentations and Contrastive Learning
• Distance Transform Guided Mixup for Alzheimer's Detection
• LangDAug: Langevin Data Augmentation for Multi-Source Domain Generalization in Medical Image Segmentation
• Continuous Domain Generalization
• Multi-Source Collaborative Style Augmentation and Domain-Invariant Learning for Federated Domain Generalization
• Graph-Based Adversarial Domain Generalization with Anatomical Correlation Knowledge for Cross-User Human Activity Recognition
• Calibrated Feature Decomposition for Generalizable Person Re-Identification
• Token-Level Prompt Mixture with Parameter-Free Routing for Federated Domain Generalization
• Improving Generalization in MRI-Based Deep Learning Models for Total Knee Replacement Prediction
• FrogDogNet: Fourier frequency Retained visual prompt Output Guidance for Domain Generalization of CLIP in Remote Sensing
• CLIP-Powered Domain Generalization and Domain Adaptation: A Comprehensive Survey
• Vision and Language Integration for Domain Generalization
• Bayesian Cross-Modal Alignment Learning for Few-Shot Out-of-Distribution Generalization
• DGFamba: Learning Flow Factorized State Space for Visual Domain Generalization
• Decentralized Domain Generalization with Style Sharing: Formal Model and Convergence Analysis
• MOSAIC: Maximizing out-of-distribution sensitivity via aligned image classification
• Generative Classifier for Domain Generalization

Related

• CNNs with Multi-Level Attention for Domain Generalization
• Learning to Generalize One Sample at a Time with Self-Supervision
• Learning Robust Representations by Projecting Superficial Statistics Out
• Improving out-of-distribution generalization via multi-task self-supervised pretraining
• Deep Domain Generalization with Feature-norm Network