Reference-Guided Machine Unlearning
基于参考的机器去学习
机构 * Department of Computer Science, ETH Zurich(苏黎世联邦理工学院计算机科学系)
AI总结 本文提出ReGUn框架,通过参考引导蒸馏实现分布不可区分性,提升去学习的遗忘-实用性权衡性能。
Comments 12 pages, 1 figure, 4 tables. Accepted at three ICLR 2026 workshops: Test-Time Updates (TTU), AI with Recursive Self-Improvement (RSI), and Agents in the Wild (AIWILD)