Resisting Large Data Variations via Introspective Transformation Network
This addresses the challenge of building accurate and robust image classifiers for computer vision applications, though it appears incremental as it builds upon existing introspective networks and data augmentation strategies.
The paper tackles the problem of training deep networks to generalize across large variations between training and testing data by proposing an introspective transformation network that embeds a learnable transformation module, achieving significant classification accuracy improvements on benchmark datasets such as MNIST, affNIST, SVHN, CIFAR-10, and miniImageNet.
Training deep networks that generalize to a wide range of variations in test data is essential to building accurate and robust image classifiers. One standard strategy is to apply data augmentation to synthetically enlarge the training set. However, data augmentation is essentially a brute-force method which generates uniform samples from some pre-defined set of transformations. In this paper, we propose a principled approach to train networks with significantly improved resistance to large variations between training and testing data. This is achieved by embedding a learnable transformation module into the introspective network, which is a convolutional neural network (CNN) classifier empowered with generative capabilities. Our approach alternates between synthesizing pseudo-negative samples and transformed positive examples based on the current model, and optimizing model predictions on these synthesized samples. Experimental results verify that our approach significantly improves the ability of deep networks to resist large variations between training and testing data and achieves classification accuracy improvements on several benchmark datasets, including MNIST, affNIST, SVHN, CIFAR-10 and miniImageNet.