An Evaluation of Lightweight CNNs for Smart Contract Vulnerability Detection

Abstract

The proposed work investigates the use of lightweight convolutional neural networks (CNNs) for detecting vulnerability patterns in Solidity RGB-encoded smart contracts. Unlike heavy CNN models, which can be computationally intensive and fall short of optimal accuracy levels, the proposed study emphasizes efficiency. Transforming smart contract source code into RGB images not only reinforces security and protects proprietary information but also addresses compactness concerns, enabling convenient storage on online platforms. This approach ensures efficient use of bandwidth, enabling rapid scanning of contracts for potential vulnerabilities post-deployment. The streamlined mechanism allows for quick and simultaneous assessment of thousands of contracts within seconds, a task that proves challenging with rigorous formal verification tools. This methodology aligns with the need for both security and efficiency in the dynamic landscape of smart contract development and deployment.