Google’s has created method for automatic image augmentation using Deep Reinforcement Learning:
Sounds pretty interesting:
“… on CIFAR10 we achieve an error rate of 1.48%, which is a 0.83% improvement over the default data augmentation designed by scientists.”