MLFuzzer: a fuzzing approach based on generative adversarial networks and BiLSTM for detecting vulnerabilities in smart contracts

MLFuzzer: a fuzzing approach based on generative adversarial networks and BiLSTM for detecting vulnerabilities in smart contracts

연구 분야: Strategies

학회: Cluster Computing

Smart contracts, composed of many programming languages, have become increasingly popular. However, they are susceptible to logical defects and security risks, which can result in financial damages and compromise the integrity of the blockchain. This work aims to use machine-learning techniques to automate the production of inputs for fuzzing, specifically for generation-based fuzzing. to be more precise, we utilize the generative adversarial network (GAN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. Our approach utilizes Generative Adversarial Networks (GANs) to produce guided inputs that are not only realistic but also highly likely to exploit vulnerabilities. This technique enhances the effectiveness of fuzzing by increasing its efficiency. The inputs, including those produced by conventional fuzzing approaches, are subsequently inputted into a BiLSTM model that has been trained using labeled data to forecast their vulnerability potential. We evaluated our approach using publicly available data sets named Smartbugs-wild, assessing the performance of MLFUZZ compared with existing state-of-the-art fuzzing tools. MLFUZZ efficiently utilizes its test inputs to test 86.2% of the time compared to SMARTGIFT. The evidence substantiates that MLFUZZ has the ability to proficiently generate test cases for the purpose of fuzzing smart contracts.