AutoAugment is a method for automatically learning data augmentation policies to improve the accuracy of modern image classifiers. The authors describe a search space where a policy consists of multiple sub-policies, each containing two image processing operations (e.g., translation, rotation, or shearing) with associated probabilities and magnitudes. A reinforcement learning algorithm is used to find the best policy that maximizes validation accuracy on a target dataset. The method achieves state-of-the-art accuracy on several datasets, including CIFAR-10, CIFAR-100, SVHN, and ImageNet, without additional data. The learned augmentation policies are also transferable to other datasets, showing significant improvements on various fine-grained classification tasks. The paper discusses the effectiveness of AutoAugment, compares it to other automated data augmentation methods, and provides ablation experiments to validate its key components.AutoAugment is a method for automatically learning data augmentation policies to improve the accuracy of modern image classifiers. The authors describe a search space where a policy consists of multiple sub-policies, each containing two image processing operations (e.g., translation, rotation, or shearing) with associated probabilities and magnitudes. A reinforcement learning algorithm is used to find the best policy that maximizes validation accuracy on a target dataset. The method achieves state-of-the-art accuracy on several datasets, including CIFAR-10, CIFAR-100, SVHN, and ImageNet, without additional data. The learned augmentation policies are also transferable to other datasets, showing significant improvements on various fine-grained classification tasks. The paper discusses the effectiveness of AutoAugment, compares it to other automated data augmentation methods, and provides ablation experiments to validate its key components.