In this paper, we investigate the use of deep learning techniques to identify and classify smart contract code vulnerabilities. We collected a large-scale dataset of smart contracts that we used to train different Convolutional Neural Networks (CNNs) models. In particular, we used two variants of 2-dimensional CNNs working on RGB images corresponding to contract byte-code, a 1-dimensional CNN working on the bytecode directly, and a Long Short-Term Memory (LSTM) neural network. Given a set of vulnerability detectors, we employed five classes of vulnerabilities. Our results show that CNNs provide a good level of accuracy and demonstrate the viability of using deep learning techniques to identify smart contract vulnerabilities.
Rossini M., Zichichi M., Ferretti S. (2023). On the Use of Deep Neural Networks for Security Vulnerabilities Detection in Smart Contracts. Institute of Electrical and Electronics Engineers Inc. [10.1109/PerComWorkshops56833.2023.10150302].
On the Use of Deep Neural Networks for Security Vulnerabilities Detection in Smart Contracts
Zichichi M.;Ferretti S.
2023
Abstract
In this paper, we investigate the use of deep learning techniques to identify and classify smart contract code vulnerabilities. We collected a large-scale dataset of smart contracts that we used to train different Convolutional Neural Networks (CNNs) models. In particular, we used two variants of 2-dimensional CNNs working on RGB images corresponding to contract byte-code, a 1-dimensional CNN working on the bytecode directly, and a Long Short-Term Memory (LSTM) neural network. Given a set of vulnerability detectors, we employed five classes of vulnerabilities. Our results show that CNNs provide a good level of accuracy and demonstrate the viability of using deep learning techniques to identify smart contract vulnerabilities.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
