原型SAE:用于大型视觉模型概念提取的自适应且稳定的字典学习
Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models
AI总结:
针对稀疏自编码器字典不稳定的问题,提出原型SAE(A-SAE)及其松弛变体RA-SAE,通过将字典原子约束在数据凸包内增强稳定性,并引入两个基准验证其重建能力与概念提取质量。
摘要:
稀疏自编码器(SAEs)已成为机器学习可解释性的强大框架,能够将模型表示无监督地分解为抽象、人类可解释概念的字典。然而,我们揭示了一个根本性局限:现有SAEs表现出严重的不稳定性,因为相似数据集上训练的相同模型可能产生截然不同的字典,这削弱了其作为可解释性工具的可靠性。为解决此问题,我们从Cutler & Breiman(1994)提出的原型分析框架中汲取灵感,提出了原型SAEs(A-SAE),其中字典原子被约束在数据的凸包内。这种几何锚定显著增强了推断字典的稳定性,而其轻度松弛变体RA-SAEs进一步匹配了最先进的重建能力。为严格评估SAEs学习的字典质量,我们引入了两个新基准,测试(i)合理性,即字典是否恢复“真实”的分类方向,以及(ii)可辨识性,即字典是否解开合成概念混合物。在所有评估中,RA-SAEs始终产生更结构化的表示,同时在大型视觉模型中发现新颖、语义上有意义的概念。
英文摘要:
Sparse Autoencoders (SAEs) have emerged as a powerful framework for machine learning interpretability, enabling the unsupervised decomposition of model representations into a dictionary of abstract, human-interpretable concepts. However, we reveal a fundamental limitation: existing SAEs exhibit severe instability, as identical models trained on similar datasets can produce sharply different dictionaries, undermining their reliability as an interpretability tool. To address this issue, we draw inspiration from the Archetypal Analysis framework introduced by Cutler & Breiman (1994) and present Archetypal SAEs (A-SAE), wherein dictionary atoms are constrained to the convex hull of data. This geometric anchoring significantly enhances the stability of inferred dictionaries, and their mildly relaxed variants RA-SAEs further match state-of-the-art reconstruction abilities. To rigorously assess dictionary quality learned by SAEs, we introduce two new benchmarks that test (i) plausibility, if dictionaries recover "true" classification directions and (ii) identifiability, if dictionaries disentangle synthetic concept mixtures. Across all evaluations, RA-SAEs consistently yield more structured representations while uncovering novel, semantically meaningful concepts in large-scale vision models.
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。