← back to paper
arxiv: 2605.01735 · 2 revisions
Less is More: Geometric Unlearning for LLMs with Minimal Data Disclosure