DeepCoFFEA: Improved Flow Correlation Attacks on Tor via Metric Learning and Amplification
Published in IEEE Symposium on Security and Privacy (SP), 2022
DeepCoFFEA introduces a novel approach to flow correlation attacks on the Tor network. The technique combines deep metric learning, which trains two feature embedding networks to map Tor and exit flows into a low-dimensional space, with an amplification strategy that divides flows into short windows and aggregates the results. DeepCoFFEA significantly reduces the number of false positives, improving the effectiveness of flow correlation attacks. Our experiments show that DeepCoFFEA achieves a true positive rate of 93% compared to only 13% in previous state-of-the-art attacks, with two orders of magnitude speedup in computational cost. This work highlights the urgent need for new traffic analysis defenses in anonymity networks like Tor.
Recommended citation: Se Eun Oh, Taiji Yang, Nate Mathews, James K. Holland, Mohammad Saidur Rahman, Nicholas Hopper, Matthew Wright. (2022). "DeepCoFFEA: Improved Flow Correlation Attacks on Tor via Metric Learning and Amplification." IEEE Symposium on Security and Privacy (SP), 2022. DOI: 10.1109/SP46214.2022.9833801.
Download Paper