Generalizing to Unseen Domains via Adversarial Data Augmentation

Abstract

We are concerned with learning models that generalize well to different \emph{unseen} domains. We consider a worst-case formulation over data distributions that are near the source domain in the feature space. Only using training data from a single source distribution, we propose an iterative procedure that augments the dataset with examples from a fictitious target domain that is "hard" under the current model. We show that our iterative scheme is an adaptive data augmentation method where we append adversarial examples at each iteration. For softmax losses, we show that our method is a data-dependent regularization scheme that behaves differently from classical regularizers that regularize towards zero (e.g., ridge or lasso). On digit recognition and semantic segmentation tasks, our method learns models improve performance across a range of a priori unknown target domains.

Accepted to NIPS 2018 (camera ready)

References in corpus (7)

• Adam: A Method for Stochastic Optimization
• A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
• Fully Convolutional Networks for Semantic Segmentation
• FCNs in the Wild: Pixel-level Adversarial and Constraint-based Adaptation
• Domain Generalization via Invariant Feature Representation
• Adversarial Discriminative Domain Adaptation
• Certifying Some Distributional Robustness with Principled Adversarial Training

Cited by in corpus (42)

• Towards Out-Of-Distribution Generalization: A Survey
• Revealing the Invisible with Model and Data Shrinking for Composite-database Micro-expression Recognition
• Test-Time Adaptable Neural Networks for Robust Medical Image Segmentation
• AADG: Automatic Augmentation for Domain Generalization on Retinal Image Segmentation
• Data synthesis and adversarial networks: A review and meta-analysis in cancer imaging
• Semantic Robustness of Models of Source Code
• Domain Generalization via Optimal Transport with Metric Similarity Learning
• 3D-VField: Adversarial Augmentation of Point Clouds for Domain Generalization in 3D Object Detection
• OSLNet: Deep Small-Sample Classification with an Orthogonal Softmax Layer
• Self-Challenging Improves Cross-Domain Generalization
• Improving Generalizability of Graph Anomaly Detection Models via Data Augmentation
• Sample Out-Of-Sample Inference Based on Wasserstein Distance
• Generalized Resilience and Robust Statistics
• Optimal Transport Based Distributionally Robust Optimization: Structural Properties and Iterative Schemes
• On the effectiveness of adversarial training against common corruptions
• Video Unsupervised Domain Adaptation with Deep Learning: A Comprehensive Survey
• Explainable AI: XAI-Guided Context-Aware Data Augmentation
• Semi-supervised Learning based on Distributionally Robust Optimization
• Variational Disentanglement for Domain Generalization
• Confidence Regions in Wasserstein Distributionally Robust Estimation
• Unsupervised Domain Adaptation in the Absence of Source Data
• An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation
• Mosaicking to Distill: Knowledge Distillation from Out-of-Domain Data
• Model-Based Domain Generalization
• Generalization of Reinforcement Learning with Policy-Aware Adversarial Data Augmentation
• 3D Adversarial Augmentations for Robust Out-of-Domain Predictions
• Beyond Domain Adaptation: Unseen Domain Encapsulation via Universal Non-volume Preserving Models
• Self-balanced Learning For Domain Generalization
• Domain generalization of 3D semantic segmentation in autonomous driving
• Robustness through Data Augmentation Loss Consistency
• Causality-inspired Single-source Domain Generalization for Medical Image Segmentation
• Embracing the Dark Knowledge: Domain Generalization Using Regularized Knowledge Distillation
• Structured Latent Embeddings for Recognizing Unseen Classes in Unseen Domains
• Generalizing Nucleus Recognition Model in Multi-source Images via Pruning
• Adapting to Unseen Vendor Domains for MRI Lesion Segmentation
• Contrastive ACE: Domain Generalization Through Alignment of Causal Mechanisms
• A Probabilistic Framework for Knowledge Graph Data Augmentation
• Learning Less Generalizable Patterns with an Asymmetrically Trained Double Classifier for Better Test-Time Adaptation
• Collaborating Foundation Models for Domain Generalized Semantic Segmentation
• Improved Detection of Face Presentation Attacks Using Image Decomposition
• Towards Data-Free Domain Generalization
• Human Imperceptible Attacks and Applications to Improve Fairness