Contrastive Learning with Adversarial Examples

Abstract

Contrastive learning (CL) is a popular technique for self-supervised learning (SSL) of visual representations. It uses pairs of augmentations of unlabeled training examples to define a classification task for pretext learning of a deep embedding. Despite extensive works in augmentation procedures, prior works do not address the selection of challenging negative pairs, as images within a sampled batch are treated independently. This paper addresses the problem, by introducing a new family of adversarial examples for constrastive learning and using these examples to define a new adversarial training algorithm for SSL, denoted as CLAE. When compared to standard CL, the use of adversarial examples creates more challenging positive pairs and adversarial training produces harder negative pairs by accounting for all images in a batch during the optimization. CLAE is compatible with many CL methods in the literature. Experiments show that it improves the performance of several existing CL baselines on multiple datasets.

arXiv commentNeurIPS 2020

References in corpus (29)

• A Simple Framework for Contrastive Learning of Visual Representations
• Fine-Grained Visual Classification of Aircraft
• Momentum Contrast for Unsupervised Visual Representation Learning
• Theoretically Principled Trade-off between Robustness and Accuracy
• Learning multiple visual domains with residual adapters
• Contrastive Multiview Coding
• Fast is better than free: Revisiting adversarial training
• Adversarial Attacks and Defences: A Survey
• Adversarial Machine Learning at Scale
• Robustness May Be at Odds with Accuracy
• On Mutual Information Maximization for Representation Learning
• Adversarial Examples: Attacks and Defenses for Deep Learning
• Self-supervised Visual Feature Learning with Deep Neural Networks: A Survey
• Unsupervised Feature Learning via Non-Parametric Instance-level Discrimination
• Unsupervised Feature Extraction by Time-Contrastive Learning and Nonlinear ICA
• Improvements to context based self-supervised learning
• Adversarial Training and Robustness for Multiple Perturbations
• Universal representations:The missing link between faces, text, planktons, and cat breeds
• Supervised Contrastive Learning
• DeepFool: a simple and accurate method to fool deep neural networks
• Learning Representations for Automatic Colorization
• Adversarial Examples for Semantic Segmentation and Object Detection
• Universal adversarial perturbations
• Unsupervised Embedding Learning via Invariant and Spreading Instance Feature
• Adversarial Examples Improve Image Recognition
• Smart Mining for Deep Metric Learning
• Learning Image Representations by Completing Damaged Jigsaw Puzzles
• Adversarial Contrastive Estimation
• Video Jigsaw: Unsupervised Learning of Spatiotemporal Context for Video Action Recognition

Cited by in corpus (6)

• SimGRACE: A Simple Framework for Graph Contrastive Learning without Data Augmentation
• Learning Representations from Audio-Visual Spatial Alignment
• Understand and Improve Contrastive Learning Methods for Visual Representation: A Review
• DimCL: Dimensional Contrastive Learning For Improving Self-Supervised Learning
• Robust Training of Temporal GNNs using Nearest Neighbours based Hard Negatives
• THAT: Two Head Adversarial Training for Improving Robustness at Scale