Smarter Contracts – Vulnerability Detection using Deep Neural Networks

Smart Contracts are computer programs that execute on a blockchain. The nature of blockchains allows one to run Smart Contracts in a trustless and decentralized environment. While different projects implement the concept of Smart Contracts, we concentrate on EVM-based blockchains and use Ethereum as our primary example, as it is the most popular, adopted, and advanced implementation.

At first glance, those Smart Contracts seem rather abstract – code running in a VM on a blockchain. However, they provide the underlying technology in a vast and fast-growing ecosystem of NFTs, decentralized applications, and – of course – CryptoKitties. All of those systems have an invested interest of million and even billion dollars. Ethereum itself has a market cap of over 250 billion USD. Furthermore, all of those systems use the fundamental promise of trustless execution, where no trusted 3rd parties are needed to establish trust between two strangers on the Internet.

The goal of this project is to demonstrate the effectiveness of Deep Neural Networks in the domain of Smart Contract vulnerability detection. Specifically, we propose to use Transfer Learning to enable the extensibility of our Machine Learning model in regards to vulnerability classes. Moreover, we show the clear benefit of Transfer Learning by successfully classifying even underrepresented vulnerability classes.

You can find more information about this project and publications on this website.

People involved: Prof. A. Dmitrienko,  Christoph Sendner