DADA: Differentiable Automatic Data Augmentation

Abstract

Data augmentation (DA) techniques aim to increase data variability, and thus train deep networks with better generalisation. The pioneering AutoAugment automated the search for optimal DA policies with reinforcement learning. However, AutoAugment is extremely computationally expensive, limiting its wide applicability. Followup works such as Population Based Augmentation (PBA) and Fast AutoAugment improved efficiency, but their optimization speed remains a bottleneck. In this paper, we propose Differentiable Automatic Data Augmentation (DADA) which dramatically reduces the cost. DADA relaxes the discrete DA policy selection to a differentiable optimization problem via Gumbel-Softmax. In addition, we introduce an unbiased gradient estimator, RELAX, leading to an efficient and effective one-pass optimization strategy to learn an efficient and accurate DA policy. We conduct extensive experiments on CIFAR-10, CIFAR-100, SVHN, and ImageNet datasets. Furthermore, we demonstrate the value of Auto DA in pre-training for downstream detection problems. Results show our DADA is at least one order of magnitude faster than the state-of-the-art while achieving very comparable accuracy. The code is available at https://github.com/VDIGPKU/DADA.

References in corpus (11)

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• Categorical Reparameterization with Gumbel-Softmax
• Improved Regularization of Convolutional Neural Networks with Cutout
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• DARTS: Differentiable Architecture Search
• The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
• MMDetection: Open MMLab Detection Toolbox and Benchmark
• Data Augmentation by Pairing Samples for Images Classification
• ShakeDrop Regularization for Deep Residual Learning
• Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules
• Monte Carlo Gradient Estimation in Machine Learning

Cited by in corpus (11)

• AutoML: A Survey of the State-of-the-Art
• Automated data processing and feature engineering for deep learning and big data applications: a survey
• An overview of mixing augmentation methods and augmentation strategies
• ResizeMix: Mixing Data with Preserved Object Information and True Labels
• Learning Invariances in Neural Networks
• Emergent Properties of Foveated Perceptual Systems
• DAAS: Differentiable Architecture and Augmentation Policy Search
• LA3: Efficient Label-Aware AutoAugment
• DIVA: Dataset Derivative of a Learning Task
• Rotating spiders and reflecting dogs: a class conditional approach to learning data augmentation distributions
• Text AutoAugment: Learning Compositional Augmentation Policy for Text Classification