让稀疏自编码器特征有序稳定，提升模型解释一致性。
Enforcing Orderedness to Improve Feature Consistency
- 通过严格排序隐空间特征，避免特征混淆。
- 在Gemma2-2B和Pythia-70M上特征一致性显著提升。
- 适合需要稳定可解释性的神经网络分析场景。
稀疏自编码器（SAEs）广泛用于神经网络的可解释性研究，但其学习到的特征在不同随机种子和超参数设置下常出现不一致。本文提出有序稀疏自编码器（OSAE），在马特里什卡自编码器基础上，通过建立严格的隐特征排序机制，并确定性地使用所有特征维度，避免了以往嵌套方法依赖采样的近似处理。理论上，我们证明在稀疏字典学习中，当解唯一时（仅自然对称性允许），OSAE可解决排列不可识别性问题。在Gemma2-2B和Pythia-70M上的实验证明，相较于马特里什卡基线，OSAE能有效提升特征一致性。
原文摘要 · Abstract (English)
Sparse autoencoders (SAEs) have been widely used for interpretability of neural networks, but their learned features often vary across seeds and hyperparameter settings. We introduce Ordered Sparse Autoencoders (OSAE), which extend Matryoshka SAEs by (1) establishing a strict ordering of latent features and (2) deterministically using every feature dimension, avoiding the sampling-based approximations of prior nested SAE methods. Theoretically, we show that OSAEs resolve permutation non-identifiability in settings of sparse dictionary learning where solutions are unique (up to natural symmetries). Empirically on Gemma2-2B and Pythia-70M, we show that OSAEs can help improve consistency compared to Matryoshka baselines.
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。