The More Popular, The Harder to Forget: Adaptive Popularity for LLM Unlearning

  • 类型:arxiv
  • 标识:2608.14229
  • 链接:https://arxiv.org/abs/2608.14229
  • 主分类:engineering
  • 形态:benchmark
  • 被引:0
  • 被引来源:Semantic Scholar
  • S2被引:0
  • 影响力被引:0
  • TLDR:The AdaPop (Adaptive Popularity) method is proposed, which combines local token confidence with a per-fact popularity-dependent exponent derived from an external proxy, and automates the forget-retain balance via a dual-ascent controller that adjusts the retain penalty each epoch.
  • 待LLM分类:否
  • 标题中文:越流行越难遗忘:面向 LLM 遗忘的自适应流行度方法
  • TLDR中文:提出 AdaPop(自适应流行度)方法,将局部 token 置信度与源自外部代理的逐事实流行度相关指数相结合,并通过双上升控制器在每个 epoch 调整 retain 惩罚来自动平衡遗忘与保留。
  • 来源文件
  • /inbox/tom/_candidates/2026-08-20-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-08-21-agent-rag-longcontext-candidates.json
  • [S2 enrich]