arXiv：cs.LG· Joshua Maccoby, Wendy Hui Wang·· 13 小时前
REDaCT：让大型推理模型免于推理崩塌的遗忘方法
REDaCT: Collapse-Free Unlearning in Large Reasoning Models
AI 导读
REDaCT 是一种面向大型推理模型（LRM）的推理时遗忘框架，通过二元泄露检测器定位推理链中泄露遗忘集信息的句子，再在推理时注入离线学习的纠错嵌入向量，引导后续推理避开泄露内容。实验显示其完全避免推理崩塌，在保持答案近乎完美遗忘与 CoT 近乎完美遗忘的同时，推理能力保留显著优于 SOTA 遗忘方法。代码已开源，论文被 EMNLP 2026 Findings 接收。
正文
Abstract:Unlearning in large reasoning models (LRMs) requires removing forget-set information from both the reasoning trace and the final answer. Existing approaches achieve this objective at the cost of reasoning collapse, where the model fails to produce a coherent chain-of-thought (CoT) and does not properly terminate its reasoning process. To address this challenge, we propose REDaCT (REgenerate via Detection and Corrective Tuning), an inference-time unlearning framework for LRMs. REDaCT consists of two key components: (1) a binary leak detector that identifies sentences in the reasoning trace that leak forget-set information, and (2) an embedding optimization module that learns a corrective embedding offline, such that when injected at the first detected leak during inference, it guides subsequent reasoning towards sentences that do not leak forget-set information. Experimental results show that REDaCT completely avoids reasoning collapse, substantially improving reasoning preservation compared to state-of-the-art unlearning methods while maintaining perfect answer forgetting and near-perfect CoT forgetting. Our code is available at: this https URL.
| Comments: | Accepted to Findings of EMNLP 2026. 19 pages, 7 figures, 6 tables. Copyright 2026 Association for Computational Linguistics (ACL). Licensed under CC BY 4.0 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.12769 [cs.LG] |
| (or arXiv:2610.12769v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12769 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Joshua Maccoby [view email]
[v1]
Thu, 8 Oct 2026 22:22:02 UTC (429 KB)
来源：arXiv：cs.LG · arxiv.org