AutoAugment is a method for automatically learning data augmentation policies to improve the accuracy of image classifiers. The approach involves searching for optimal augmentation strategies through a discrete search space, where each policy consists of multiple sub-policies. Each sub-policy includes two image processing operations, along with their probabilities and magnitudes. The search algorithm, based on reinforcement learning, finds the best policy that maximizes validation accuracy on a target dataset. AutoAugment achieves state-of-the-art results on several datasets, including CIFAR-10, CIFAR-100, SVHN, and ImageNet. On ImageNet, it achieves a Top-1 accuracy of 83.5%, surpassing previous records. The policies learned on ImageNet can be transferred to other datasets, such as Oxford Flowers, Caltech-101, Oxford-IIT Pets, FGVC Aircraft, and Stanford Cars, leading to significant improvements. The method is effective in both direct application to a dataset and transfer to new datasets. AutoAugment outperforms other automated data augmentation methods, demonstrating its effectiveness in improving generalization accuracy. The search space includes 16 image processing operations, and the policies are found through a combination of reinforcement learning and policy gradient methods. The results show that AutoAugment is a powerful tool for improving image classification performance through data augmentation.AutoAugment is a method for automatically learning data augmentation policies to improve the accuracy of image classifiers. The approach involves searching for optimal augmentation strategies through a discrete search space, where each policy consists of multiple sub-policies. Each sub-policy includes two image processing operations, along with their probabilities and magnitudes. The search algorithm, based on reinforcement learning, finds the best policy that maximizes validation accuracy on a target dataset. AutoAugment achieves state-of-the-art results on several datasets, including CIFAR-10, CIFAR-100, SVHN, and ImageNet. On ImageNet, it achieves a Top-1 accuracy of 83.5%, surpassing previous records. The policies learned on ImageNet can be transferred to other datasets, such as Oxford Flowers, Caltech-101, Oxford-IIT Pets, FGVC Aircraft, and Stanford Cars, leading to significant improvements. The method is effective in both direct application to a dataset and transfer to new datasets. AutoAugment outperforms other automated data augmentation methods, demonstrating its effectiveness in improving generalization accuracy. The search space includes 16 image processing operations, and the policies are found through a combination of reinforcement learning and policy gradient methods. The results show that AutoAugment is a powerful tool for improving image classification performance through data augmentation.