This repository consists of scripts for the paper Can Machine Unlearning Reduce Social Bias in Language Models? [To be published to EMNLP 2024 Industry Track].
- Install required packages:
python -m pip install -r requirements.txt- Instructions for running scripts are available in the respective directories for each method.
This work has resulted from a larger collaborative initiative involving the Vector Institute and its industry partners. The authors extend their appreciation to Tahniat Khan, the project manager, for her efforts in coordinating this project. We also express our thanks to Deval Pandya, Vice President of AI Engineering at the Vector Institute, for his valuable support.
The authors would like to acknowledge the leaders at Ernst & Young (EY) for their exceptional support and commitment to advancing artificial intelligence research. Special thanks to Mario Schlener, Managing Partner for Risk Consulting Canada, whose strategic vision exemplifies EY's dedication to fostering innovation and thought leadership in the industry. We also recognize the expert oversight of Yara Elias, Kiranjot Dhillon, and Rasoul Shahsavarifar from AI Risk Canada, whose contributions were integral to the project's success. This partnership not only reflects EY's investment in AI but also sets a foundation for continued research collaboration and driving progress in the field.