从权重到特征：SAE引导的激活正则化用于大语言模型持续学习
From Weights to Features: SAE-Guided Activation Regularization for LLM Continual Learning
AI总结：
针对大语言模型持续学习中的灾难性遗忘问题，提出利用预训练稀疏自编码器（SAE）在激活空间进行正则化，通过特征掩码平衡稳定性和可塑性，无需旧任务数据，内存效率高，在TRACE和MedCL基准上超越EWC等方法。
摘要：
权重空间正则化方法如弹性权重巩固（EWC）是持续学习中应对灾难性遗忘的标准方法。然而，这些方法应用于大语言模型时往往表现不佳。我们认为这种不佳表现部分可归因于大语言模型的“多语义”特性：EWC风格正则化使用的每权重重要性估计过于粗糙，无法隔离需要保护的知识。本文提出在模型的激活空间中进行正则化，使用预训练的稀疏自编码器（SAE）作为单语义特征字典。从约束优化的角度，我们推导出一个新的损失函数，利用SAE特征字典显式平衡稳定性和可塑性，并证明EWC是单侧权重空间惩罚设置中的一个特例。与基于回放的方法（存储或重访早期任务示例）不同，我们的方法在掩码构建后不需要先前任务数据：当前任务数据用于计算紧凑的SAE特征掩码，仅保留该掩码用于后续训练。此外，由于特征空间的维度远低于参数空间，所提方法内存效率更高。在TRACE和MedCL持续学习基准上，该方法在不引入任务特定架构组件的方法中取得了最强结果，也超越了EWC等传统权重空间正则化方法。除了性能比较，我们为多语义性论点提供了实证证据：任务相关表示在SAE特征基中是线性可分的，但在权重基中与随机无异，且权重空间保护在概念层面上几乎是非选择性的。
英文摘要：
Weight-space regularization methods such as Elastic Weight Consolidation (EWC) are the standard approach to catastrophic forgetting in continual learning. However, those methods tend to underperform when applied to large language models. We argue that such underperformance can be partly explained by the ``polysemantic'' nature of large language models: per-weight importance estimates utilized by EWC-style regularization are too coarse and cannot isolate the knowledge that needs protection. In this paper, we propose regularizing instead in the model's activation space, using pretrained Sparse Autoencoders (SAEs) as a monosemantic feature dictionary. From the perspective of constrained optimization, we derive a new loss function that uses the SAE feature dictionary to explicitly balance stability and plasticity, and show that EWC is a special case in the one-sided weight-space penalty setting. Unlike replay-based methods that store or revisit examples from earlier tasks, our method requires no previous-task data after mask construction: current-task data is used to compute a compact SAE feature mask, and only this mask is retained for later training. Further, since the feature space has significantly lower dimensionality than the parameter space, the proposed method is more memory efficient. On the TRACE and MedCL continual learning benchmarks, the method achieves the strongest result among approaches without introducing task-specific architectural components, also surpassing traditional weight-space regularization methods like EWC. Beyond performance comparisons, we provide empirical evidence for the polysemanticity thesis: task-relevant representations are linearly separable in the SAE feature basis but indistinguishable from chance in the weight basis, and weight-space protection is nearly non-selective at the concept level.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
