[Paper Review] ESCORT: Ethereum Smart COntRacTs Vulnerability Detection using Deep Neural Network and Transfer Learning

O. Lutz, Huili Chen+5 more|arXiv (Cornell University)|Mar 23, 2021

Blockchain Technology Applications and Security53 references32 citations

TL;DR

ESCORT introduces a multi-output DNN framework that detects multiple Ethereum smart contract vulnerabilities from bytecode, with transfer learning to add new vulnerability types efficiently.

ABSTRACT

Ethereum smart contracts are automated decentralized applications on the blockchain that describe the terms of the agreement between buyers and sellers, reducing the need for trusted intermediaries and arbitration. However, the deployment of smart contracts introduces new attack vectors into the cryptocurrency systems. In particular, programming flaws in smart contracts can be and have already been exploited to gain enormous financial profits. It is thus an emerging yet crucial issue to detect vulnerabilities of different classes in contracts in an efficient manner. Existing machine learning-based vulnerability detection methods are limited and only inspect whether the smart contract is vulnerable, or train individual classifiers for each specific vulnerability, or demonstrate multi-class vulnerability detection without extensibility consideration. To overcome the scalability and generalization limitations of existing works, we propose ESCORT, the first Deep Neural Network (DNN)-based vulnerability detection framework for Ethereum smart contracts that support lightweight transfer learning on unseen security vulnerabilities, thus is extensible and generalizable. ESCORT leverages a multi-output NN architecture that consists of two parts: (i) A common feature extractor that learns the semantics of the input contract; (ii) Multiple branch structures where each branch learns a specific vulnerability type based on features obtained from the feature extractor. Experimental results show that ESCORT achieves an average F1-score of 95% on six vulnerability types and the detection time is 0.02 seconds per contract. When extended to new vulnerability types, ESCORT yields an average F1-score of 93%. To the best of our knowledge, ESCORT is the first framework that enables transfer learning on new vulnerability types with minimal modification of the DNN model architecture and re-training overhead.

Motivation & Objective

• Motivation to detect multiple vulnerability types in Ethereum smart contracts efficiently.
• Develop a scalable, extensible DNN framework that operates on bytecode without source code access.
• Enable transfer learning to incorporate new vulnerability types with minimal retraining.
• Provide probabilistic outputs per vulnerability to improve interpretability.

Proposed method

• Proposes a multi-output DNN with a shared feature extractor (stem) and per-vulnerability branches (heads).
• Uses an embedding layer to process long contract bytecode and GRU/LSTM layers to capture sequence semantics.
• Each vulnerability branch outputs a sigmoid probability for its type, enabling multi-label detection.
• Introduces a transfer learning strategy that freezes the feature extractor and existing branches while adding/training a new branch for new vulnerabilities.
• Evaluation on a large bytecode dataset labels contracts using Oyente, Mythril, and Dedaub tools.
• Demonstrates 0.02 seconds per contract inference and 95% average F1 across six types; transfer learning achieves 93% average F1 for new types.

Experimental results

Research questions

• RQ1 RQ1: Can a single DL model detect multiple vulnerability types in smart contracts?
• RQ2 RQ2: Can the DL model be extended to new vulnerabilities via transfer learning with minimal retraining?
• RQ3 RQ3: How does ESCORT perform in terms of accuracy, efficiency, and extensibility compared to prior tools?

Key findings

• ESCORT achieves an average F1 score of 95% across six vulnerability types.
• Inference time is 0.02 seconds per contract.
• When extended to new vulnerability types, ESCORT yields an average F1 score of 93%.
• ESCORT is the first framework enabling transfer learning on new vulnerability types with minimal modification and retraining overhead.
• A large dataset of 93,497 contracts labeled via Oyente, Mythril, and Dedaub supports evaluation.
• The architecture enables fast extension by adding new branches to the existing multi-output model.