/unlearning-verification

verifying machine unlearning by backdooring

Primary LanguagePython

Athena: Probabilistic Verification of Machine Unlearning

This repository contains the relevant code required to reproduce the results in the paper here. This work was published at Privacy Enhancing Technologies Symposium (PETS) 2022.

Table of Contents

Requirements

  • The code should work on most Linux distributions (It has been developed and tested with Ubuntu 21.10).
  • The following packages are required: pkg-config libhdf5-dev
  • The scripts are written for python3.9, the required packages are listed in requirements.txt
  • The python3 package manager pip3 is required to be installed.

Source Code

All the source code is provided in src/ and Makefile is provided inside.

  • src/single_user: Code for single users. This is the heart of our project.
  • src/multi_user: This directory contains scripts analyzing the involvment of collaborating users.
  • src/plotting_scripts: This directory contains plotting scripts.

Docker Image

There is a Dockerfile that autmatically downloads and builds all required content and dependencies. To build and run, download the Dockerfile and execute in the downloading directory the following commands with bash:

$ docker build . -t unlearning-verification
$ docker run -it unlearning-verification /bin/bash

Then, continue with the code-running instructions below.

Installing requirements

Install required packages using the following command (For Ubuntu 21.10): sudo apt install pkg-config libhdf5-dev build-essential git python3 python3-venv

Running the code

This code runs on python3. It requires python3 and pip (with the command "pip" refering to python3, not python2) to be installed. After installing the dependencies (or if docker has done it for you), you can run the code with the following comand: cd src/; make

Each sub-director contains a README file to provide pointers on the relevant experiments.

Citation

You can cite the paper using the following bibtex entry (the paper links to this repo):

@inproceedings{sommer2022athena,
  title={{Athena: Probabilistic Verification of Machine Unlearning}},
  author={Sommer, David M. and SOng, Liwei and Wagh, Sameer and Mittal, Prateek},
  journal={Proceedings on Privacy Enhancing Technologies},
  year={2022}
}

Warning

This codebase is released solely as a reference for other developers, as a proof-of-concept, and for benchmarking purposes.

For questions, please create git issues.