Deep Learning in Fuzzing: A Literature Survey

Fuzzing is one of the most popular vulnerability discovery techniques today. Traditional fuzzing is faced with many drawbacks such as poor efficiencies and low code coverage, and expert experience largely determines the effect of fuzzing. Deep learning, which has made remarkable achievements in a variety of artificial intelligence(AI) research areas, provides promising ways for the development of fuzzing. This paper provides a comprehensive review of fuzzing techniques based on deep learning methods. First, we give an introduction to fuzzing and deep learning. Then, we introduce the application research of deep learning methods in fuzzing from the perspective of test cases and program analysis in the fuzzing process. Finally, we conclude the limitations and challenges faced in the current research of fuzzing techniques based on deep learning methods and discuss new trends and potential directions for the future development of this field.