gradient ascent

3 papers

• Ascent Fails to Forget
• Momentum-SAM: Sharpness Aware Minimization without Computational Overhead
• Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning