Abstract:
"The rapid evolution of blockchain technology has ushered in the era of smart contracts, revolutionizing decentralized agreements. However, this transformative landscape has brought forth a critical challenge – the vulnerability inherent in smart contracts, posing a threat to their reliability. This research endeavours to address these challenges by leveraging advanced machine learning techniques, with a primary focus on bolstering the security infrastructure of smart contracts. The overarching aim is to meticulously identify and rectify vulnerabilities embedded in smart contract code, fortifying their foundational integrity, and contributing to the development of resilient smart contracts capable of withstanding emergent threats.
In tackling this problem, the research adopts a novel approach that eschews traditional detection methods. Instead, it employs a sophisticated system integrating Graph Neural Networks (GNN) with Call Graphs (CG) and Control Flow Graphs (CFG). This innovative methodology allows for the comprehensive and precise detection of vulnerabilities within smart contract source code. The iterative prototyping process ensures validation, refinement, and user feedback, contributing to a deeper understanding of the problem domain.
Initial results demonstrate the efficacy of the proposed solution. Quantitative analyses reveal a significant improvement in the identification and localization of vulnerabilities, surpassing the limitations of conventional methods. The system not only addresses known vulnerabilities but also exhibits a remarkable capability to capture previously unknown threats, enhancing the overall security posture of smart contracts. This pioneering research marks a substantial advancement in smart contract security, laying the foundation for a paradigm shift in vulnerability detection methodologies."
