Affinity and Diversity: Quantifying Mechanisms of Data Augmentation

Abstract

Though data augmentation has become a standard component of deep neural network training, the underlying mechanism behind the effectiveness of these techniques remains poorly understood. In practice, augmentation policies are often chosen using heuristics of either distribution shift or augmentation diversity. Inspired by these, we seek to quantify how data augmentation improves model generalization. To this end, we introduce interpretable and easy-to-compute measures: Affinity and Diversity. We find that augmentation performance is predicted not by either of these alone but by jointly optimizing the two.

10 pages, 7 figures

References in corpus (26)

• Auto-Encoding Variational Bayes
• EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
• mixup: Beyond Empirical Risk Minimization
• SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
• Improved Regularization of Convolutional Neural Networks with Cutout
• Asymptotic Equivalence of Bayes Cross Validation and Widely Applicable Information Criterion in Singular Learning Theory
• FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence
• Unsupervised Data Augmentation for Consistency Training
• Understanding deep learning requires rethinking generalization
• Random Erasing Data Augmentation
• AutoAugment: Learning Augmentation Policies from Data
• EfficientDet: Scalable and Efficient Object Detection
• Data Augmentation by Pairing Samples for Images Classification
• Why do deep convolutional networks generalize so poorly to small image transformations?
• ReMixMatch: Semi-Supervised Learning with Distribution Alignment and Augmentation Anchoring
• Adversarial Examples Are a Natural Consequence of Test Error in Noise
• Further advantages of data augmentation on convolutional neural networks
• A Bayesian Data Augmentation Approach for Learning Deep Models
• APAC: Augmented PAttern Classification with Neural Networks
• Benchmarking Neural Network Robustness to Common Corruptions and Surface Variations
• StarNet: Targeted Computation for Object Detection in Point Clouds
• A Kernel Theory of Modern Data Augmentation
• Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation
• A Fourier Perspective on Model Robustness in Computer Vision
• Adversarial Examples Improve Image Recognition
• Data Augmentation Revisited: Rethinking the Distribution Gap between Clean and Augmented Data

Cited by in corpus (16)

• Hard Negative Mixing for Contrastive Learning
• FMix: Enhancing Mixed Sample Data Augmentation
• CoDA: Contrast-enhanced and Diversity-promoting Data Augmentation for Natural Language Understanding
• Anatomy of Catastrophic Forgetting: Hidden Representations and Task Semantics
• Viewmaker Networks: Learning Views for Unsupervised Representation Learning
• Selecting Data Augmentation for Simulating Interventions
• CutDepth:Edge-aware Data Augmentation in Depth Estimation
• The Deep Bootstrap Framework: Good Online Learners are Good Offline Generalizers
• KeepAugment: A Simple Information-Preserving Data Augmentation Approach
• An Empirical Analysis of the Impact of Data Augmentation on Knowledge Distillation
• SCARF: Self-Supervised Contrastive Learning using Random Feature Corruption
• ReSmooth: Detecting and Utilizing OOD Samples when Training with Data Augmentation
• Does Data Augmentation Benefit from Split BatchNorms
• A Unified Mixture-View Framework for Unsupervised Representation Learning
• Local Patch AutoAugment with Multi-Agent Collaboration
• On Data-centric Myths