Efficient fuzzing testcases generation based on SAGAN

What is it about?

Generative adversarial networks can be used for the generation of testcases in fuzzing, but the structural information of the testcases is rarely attended to. In this paper, we adopt a SAGAN-based testcases generation technique, to learn and utilize the structural information of the testcases and give attention to the important parts. We selectively improve the network structure so that the model can be more adapted to the structural information of the fuzzing testcases. We used gradient penalty and spectral normalization to stabilize the training of the network. The results show that our approach has higher efficiency on the lava-m dataset. In addition, the fuzzing testcases generated by SAGAN can find more crashes and hangs compared to those mutated by AFL++.

Featured Image

Photo by Unseen Studio on Unsplash

Photo by Unseen Studio on Unsplash

Why is it important?

This paper adopts a SAGAN-based fuzzing testcase generation to quantitatively extract the information of structure and variable byte block in fuzzing testcases automatically. This paper also designs a novel model of generative adversarial network and prototype system for quantitatively extraction of structural information which can guide generation of fuzzing testcases.