UnlearningSoup: Is Repeated Tuning Necessary for Large Language Model Unlearning?
作者: Puning Yang, Qizhou Wang, Junchi Yu, Bo Han, Xiuying Chen
分类: cs.LG
发布日期: 2026-09-29
备注: NeurIPS 2026 Accepted
💡 一句话要点
提出UnlearningSoup以解决大语言模型遗忘效率低下问题
🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 大语言模型 遗忘机制 超参数优化 机器学习 模型性能提升
📋 核心要点
- 现有的遗忘方法在平衡遗忘与知识保留时面临超参数选择困难，常需反复调优，效率低下。
- 本文提出UnlearningSoup框架，通过分析权重空间结构，减少重复调优的需求，提升模型性能。
- 实验结果显示，UnlearningSoup在超参数选择上实现了2.4到3.3倍的效率提升，且在多种设置中均表现出性能改善。
📝 摘要（中文）
大语言模型在训练过程中可能会记住有害内容，导致其输出中重新出现。现有的遗忘方法通常依赖于基于训练的参数更新，难以平衡遗忘与保留的目标，且需要反复调优。为了解决这一挑战，本文提出UnlearningSoup框架，利用权重空间中的结构性，减少重复调优的需求。该框架包含两种策略：EfficientSoup通过二分搜索插值快速发现表现良好的模型，PerformanceSoup则通过加权组合有效释放剩余性能潜力。实验表明，UnlearningSoup在超参数选择上实现了2.4到3.3倍的效率提升，并在各设置中持续改善性能。
🔬 方法详解
问题定义：本文旨在解决大语言模型在遗忘有害内容时效率低下的问题。现有方法依赖于训练参数更新，难以平衡遗忘与保留的目标，且超参数选择复杂，需反复调优。
核心思路：论文的核心思路是利用模型在权重空间中的结构性，提出UnlearningSoup框架，减少重复调优的需求。通过分析不同运行的模型在评估性能上的相似性，设计出高效的遗忘策略。
技术框架：UnlearningSoup框架包含两个主要模块：EfficientSoup和PerformanceSoup。EfficientSoup使用二分搜索插值快速找到表现良好的模型，而PerformanceSoup则在后期通过加权组合释放剩余性能潜力。
关键创新：最重要的创新点在于提出了基于权重空间结构的遗忘策略，显著减少了对超参数反复调优的依赖，与现有方法相比，提供了更高的效率和性能提升。
关键设计：在EfficientSoup中，采用二分搜索算法进行模型选择；在PerformanceSoup中，通过加权组合方法优化模型性能。具体的参数设置和损失函数设计在实验中进行了详细验证，以确保框架的有效性。
🖼️ 关键图片
📊 实验亮点
实验结果表明，UnlearningSoup在超参数选择上实现了2.4到3.3倍的效率提升，显著优于传统方法。此外，该框架在多种数据集和模型上均表现出一致的性能改善，验证了其广泛适用性和有效性。
🎯 应用场景
该研究的潜在应用领域包括大语言模型的安全性和可靠性提升，尤其是在处理敏感信息和有害内容时。通过提高遗忘效率，UnlearningSoup能够帮助开发更安全的AI系统，减少不当内容的输出，具有重要的实际价值和社会影响。未来，该框架可扩展至其他机器学习领域，提升模型的适应性和效率。
📄 摘要（原文）
Large language models trained on vast corpora inherently risk memorizing harmful content that may later re-emerge in their outputs. To mitigate this issue, existing unlearning methods typically rely on training-based parameter updates, such as gradient ascent and its variants, to delete targeted content while preserving other knowledge. However, balancing the competing goals of forgetting and retention makes hyperparameter choices for these methods particularly difficult, often requiring repeated tuning to obtain a strong model that still leaves substantial room for improvement and transfers poorly across models and datasets. To address this challenge, we investigate whether unlearning runs exhibit exploitable structure in weight space, and observe that models from different runs still lie in a shared evaluation-performance basin. This suggests that stronger models may be recovered through an unlearning-tailored soup strategy, reducing the need for repeated tuning for further improvement or new settings. Motivated by this, we propose UnlearningSoup, a unified framework that provides two strategies: EfficientSoup uses binary-search-based interpolation to quickly discover a well-performing model in the early stage, where repeated tuning would otherwise make strong model selection costly. PerformanceSoup uses reweighted souping to efficiently unlock the remaining performance potential in the later stage, where repeated tuning becomes increasingly inefficient. Extensive experiments across diverse datasets and models show that UnlearningSoup delivers 2.4x to 3.3x efficiency gains in hyperparameter selection, while consistently improving performance across settings.