TRAI: Learn to Unlearn – Towards Responsible AI (Novo Nordisk Foundation)

TRAI aims to develop Unlearning algorithms that can erase knowledge of specific data samples (raising privacy/copyright issues, or leading to biased decisions) from already trained AI models. The project primarily focuses on Convolutional Neural Networks (ConvNets) and Large Language Models (LLMs). This way, TRAI helps AI comply with GDPR’s right to be forgotten, while also promoting fairness and accountability—key aspects of responsible AI. We believe TRAI will help build trust and acceptance of AI across different fields/industries.

A big thank you to the Novo Nordisk Foundation for supporting this important project, and to The Maersk Mc-Kinney Moller Institute, Faculty of Engineering, University of Southern Denmark, for their fantastic support.