anmold-07/Model-Extraction-with-RL

ECSE 6962 - Trustworthy Machine Learning Course Project: Model Extraction attacks via Active RL policies

Model-Extraction-with-RL

ECSE 6962 - Trustworthy Machine Learning Course Project: Model Extraction attacks via Active RL policies

Requirements: under requirements.txt

To replicate results:

1. run experiments/active_learning_baseline.py to find the performance of baseline methods (both Extended Adaptive Re-training and Adaptive Re-training methods).
2. run experiments/lowd_meek.py to find the performance of the second baseline method that is based on generalized binary search.
3. run experiments/active_learning_via_RL.py to find the performance of the proposed "Active Learning via Reinforcement Learning Policies" method.

Please feel to reach out to "dwivea2" at "rpi.edu" if you have any questions or want to collaborate further on this project.

The code for the baselines are built upon the code by the original contributor and author that can be found on: https://github.com/ftramer/Steal-ML. The only caveat is that their code requires quite a few modifications in order to run with Python-3. The modified code for the baselines in this git repository gets rid of all the issues.