Text adversarial examples generation and defense based on reinforcement learning

Abstract

In recent years, the neural networks are widely used in image processing, natural language processing and other fields. But there are new security issues-the adversarial examples. Crafted adversarial examples can make a trouble for the neural network, which leads to the mis-classification. Text classification is one of the basic tasks of the natural language processing. This paper is concerned about the generation and defense of text adversarial examples. The main contributions of this research are as follows: This paper explores a new type of adversarial example and applies reinforcement learning to generate the adversarial examples; a training set composed of adversarial examples is constructed. To build a more robust classifier, a new defense framework is established. In order to eliminate the influence of noise, well-designed predetector and reformer were implemented, which helps the neural networks to resist adversarial examples and reduce coupling.