This research proposes WGAN-AFL, a method to improve the efficiency of vulnerability detection in software. The method uses fuzz testing, an automated software testing technology, to detect vulnerabilities. However, the performance of this method often depends on the quality of initial input seeds. To overcome this, the researchers use a generative adversarial network (GAN) to learn the features of high-quality test cases and generate high-quality initial input seeds. They also use the Wasserstein GAN (WGAN) architecture to address issues such as training instability and mode collapse. The experimental results show that WGAN-AFL outperforms the original AFL in terms of code coverage, new paths, and vulnerability discovery.
Publication date: 1 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.16947
