This repo provides a replicated implementation of this article DeepCorr: Strong Flow Correlation Attacks on Tor Using Deep Learning. The official implementation could be found here. The dataset is available here.
As always, the MAIN.ipynb shows the story-line. config.yml is the place in which you can tune the parameters.
The loss and accuracy history plots in MAIN.ipynb are results I've got after 200 epochs of training process with configurations pretty much the same as depicted in the article. Rather than picking up 1:199 positive-negative flows, I was playing with 1:1 positive-negative flows for all the training, val, and testing processes. (Luckily, I tried that with four Tesla V100 cards on one NVIDIA DGX-1 server.)
- Ubuntu 20.04
- Python 3
- tensorflow 2.3