Generative Model Unlearning: A Survey through Target Events, Unlearning Operators, and Evaluation Protocols
The episode surveys 'Generative Model Unlearning,' discussing how to teach AI models to forget specific data or concepts without expensive full retraining. Hosts categorize methods into point-wise and concept-wise unlearning, covering techniques from fine-tuning to inference guidance, and detailing necessary improvements for scalability and robustness.