让大模型忘记特定知识时，还能保留通用能力。
Modeling LLM Unlearning as an Asymmetric Two-Task Learning Problem
- 将遗忘任务视为辅助目标，优先保证模型能力保留。
- 新方法SAGO使模型在忘记后仍恢复96%的原始性能。
- 适合需要安全可控的AI系统开发者使用。
大语言模型的机器遗忘旨在移除特定知识的同时保持通用能力。本文将该问题重新建模为非对称双任务学习：保留是主要目标，遗忘是辅助目标。为此提出一种以保留优先的梯度合成框架，分离任务特异性梯度提取与冲突感知组合。实例化该框架，我们改进了现有PCGrad以解决梯度冲突，并提出SAGO——一种新的保留优先梯度合成方法。理论上，两种变体均确保与保留梯度的余弦相似度非负，而SAGO通过符号约束合成实现更紧密对齐。实验上，在WMDP Bio/Cyber和RWKU基准测试中，SAGO持续提升帕累托前沿表现：例如在WMDP Bio（SimNPO+GD）上，对目标模型MMLU性能的恢复从44.6%（朴素方法）提升至94.0%（+PCGrad），进一步增至96.0%（+SAGO），同时保持相当的遗忘强度。结果表明，重塑梯度几何而非平衡损失权重，才是缓解遗忘-保留权衡的关键。
原文摘要 · Abstract (English)
Machine unlearning for large language models (LLMs) aims to remove targeted knowledge while preserving general capability. In this paper, we recast LLM unlearning as an asymmetric two-task problem: retention is the primary objective and forgetting is an auxiliary. From this perspective, we propose a retention-prioritized gradient synthesis framework that decouples task-specific gradient extraction from conflict-aware combination. Instantiating the framework, we adapt established PCGrad to resolve gradient conflicts, and introduce SAGO, a novel retention-prioritized gradient synthesis method. Theoretically, both variants ensure non-negative cosine similarity with the retain gradient, while SAGO achieves strictly tighter alignment through constructive sign-constrained synthesis. Empirically, on WMDP Bio/Cyber and RWKU benchmarks, SAGO consistently pushes the Pareto frontier: e.g., on WMDP Bio (SimNPO+GD), recovery of target model MMLU performance progresses from 44.6% (naive) to 94.0% (+PCGrad) and further to 96.0% (+SAGO), while maintaining comparable forgetting strength. Our results show that re-shaping gradient geometry, rather than re-balancing losses, is the key to mitigating unlearning-retention trade-offs.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。