Machine Unlearning

Sangdon Park
Sangdon Park
GPT-5

Trust issues on Agentic AI would never end. Whenever we discover trust issues, how can we mitigate this? We leverage machine unlearning for this. To this end, we consider the following question.

How to efficiently unlearn trust issues and certify this?

On-going/Potential Projects

  • Certified Unlearning: Certify unlearning in LLMs.
  • Efficient Unlearning: Unlearn efficiently with few gradient steps.

What’s your own project idea?

Keywords

  • Machine Unlearning
  • LLMs
  • Conformal Prediction
  • Online Learning