Reinforcement Based Learning on Classification Task Yields Better Generalization and Adversarial Accuracy (Student Abstract)

Deep Learning has become interestingly popular in the field of computer vision, mostly attaining near or above human-level performance in various vision tasks. But recent work has also demonstrated that these deep neural networks are very vulnerable to adversarial examples (adversarial examples - inputs to a model which are naturally similar to original data but fools the model in classifying it into a wrong class). In this work, we proposed a novel method to train deep learning models on an image classification task. We used a reward-based optimization function, similar to the vanilla policy gradient method in reinforcement learning to train our model instead of conventional cross-entropy loss. An empirical evaluation on cifar10 dataset showed that our method outperforms the same model architecture trained using cross-entropy loss function (on adversarial training). At the same time, our method generalizes better to the training data with the difference in test accuracy and train accuracy < 2% for most of the time as compared to cross-entropy one, whose difference most of the time remains > 2%.