Machine Unlearning Research Code

This repository contains code for machine unlearning experiments, including:

1. Data poisoning and visualization
2. Hessian-based analysis of model landscapes
3. Various unlearning methods comparison (Retrain, Finetune, Scrub, Grad-Asc)
4. Evaluation of unlearning effectiveness on both classical MLP and quantum neural networks

File Structure

• datar.py: Data preprocessing and loading functions
• models.py: Model definitions for both MLP and QNN
• poison_unlearn.py: Implementation of various unlearning methods
• run_unlearn_mnist.py: Main script for MNIST unlearning experiments
• run_unlearn_xxz.py: Main script for XXZ model unlearning experiments
• test_mnist.ipynb: Jupyter notebook for MNIST experiments visualization
• test_xxz.ipynb: Jupyter notebook for XXZ model experiments visualization
• heissian.ipynb: Hessian-based analysis of model landscapes
• run_example.py: Example script demonstrating how to run the code

Requirements

• Python 3.x
• TensorFlow
• JAX
• TensorCircuit
• Optax
• Scikit-learn
• NumPy
• Matplotlib
• Seaborn

Install all requirements with:

pip install -r requirements.txt

Usage

Quick Start

Run the example script to test the code:

python run_example.py

Full Experiments

To run the MNIST unlearning experiments:

python run_unlearn_mnist.py

To run the XXZ model unlearning experiments:

python run_unlearn_xxz.py

Note: The XXZ experiments require specific dataset files that are not included in this repository.

Visualization

The results can be visualized using the provided Jupyter notebooks:

• test_mnist.ipynb: For MNIST experiment results
• test_xxz.ipynb: For XXZ model experiment results
• heissian.ipynb: For Hessian-based analysis

Methods

The repository implements several machine unlearning methods:

1. Retrain: Retrain the model from scratch on the retained data
2. Fine-tune (cf): Continue training on the retained data
3. Scrub: Use KL divergence regularization to "scrub" the influence of forgotten data
4. Gradient Ascent (ga): Use gradient ascent to actively unlearn the forgotten data

Models

Two types of models are implemented:

1. Classical MLP: Multi-layer perceptron with configurable hidden layers
2. Quantum Neural Network (QNN): Quantum circuit-based model using TensorCircuit

Data

The code supports experiments on:

1. MNIST: Binary classification of digits 1 and 9
2. XXZ Model: Quantum many-body system data (requires external dataset files)

Results

The experiments evaluate:

1. Validation Accuracy: Performance on clean validation data
2. Forgetting Accuracy: Ability to forget the "unlearned" data
3. Model Stability: Robustness of the unlearning process

License

This project is licensed under the MIT License - see the LICENSE file for details.