DeSG: Towards Generating Valid Solidity Smart Contracts with Deep Learning

【Author】 Tian, Zhenzhou; Wang, Fanfan

【Source】ADVANCES IN NATURAL COMPUTATION, FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY, ICNC-FSKD 2022

【影响因子】

【Abstract】Solidity being a young yet the most widely used programming language to develop smart contracts, which are programs that run on the blockchains, is frequently exposed of bugs and under continues improvement. To this end, this work presents DeSG to facilitate the fuzz testing of the fast-evolving Solidity compiler, by automatically generating massive solidity smart contracts of diversity in a deep learning based manner. An encoder-decoder model is designed to ensure the production of high-quality and valid smart contracts, by equipping with the powerful representation learning ability from the Transformer, as well as three carefully-designed generation strategies that well-match the features of the Solidity language. The experimental evaluations conducted show that, DeSG can effectively generate syntactically valid and highly compilable smart contracts. The impacts of different encoding neural networks and generation strategies to DeSG are also evaluated, with the best performance reaching 91.2% validity score.

【Keywords】Solidity; Smart contract; Code generation; Deep learning

【发表时间】2023

【收录时间】2023-05-10

【文献类型】实验仿真

【主题类别】

区块链技术-核心技术-智能合约