Smart Contract Reentry Vulnerability Detection Based on Federated Learning

Abstract

A federated learning based smart contract reentry vulnerability detection method is proposed to address the issues of multiple data sources and data privacy in existing smart contract reentry vulnerability detection methods, in order to address the privacy and distributed storage issues of smart contract code data. By integrating horizontal federated learning and deep learning methods, model parameters are shared among various participants without sharing raw data, achieving protection of data privacy and security. At the same time, model training is allowed on different datasets, thereby more comprehensively discovering smart contract re-entry vulnerabilities. In the detection of re-entry vulnerabilities in smart contracts, by updating model parameters locally and aggregating parameters on a central server, it is possible to effectively detect re-entry vulnerabilities in smart contracts while protecting the privacy and security of contract data. The final experiment shows that the proposed method can protect privacy while still ensuring high model accuracy.