Deep Learning for Consumer Devices and Services 4—A Review of Learnable Data Augmentation Strategies for Improved Training of Deep Neural Networks

摘要

Learnable data augmentation is a technique where a neural netowrk learns to create modified data samples that improve the training outcome from a second, parallel neural network. This is a relatively new approach to dataset augmentation that has inspired many variations in the last few years. In this article the most signficiant of these advanced data augmentation strategies are summarised and discussed.