能力 ≠ 可解释性：视觉基础模型的人类可解释性
Capability $\neq$ Interpretability: Human Interpretability of Vision Foundation Models
AI总结：
本文研究了视觉基础模型的人类可解释性，提出了一种评估框架，发现基础模型比监督模型更难解释，且可解释性与下游任务性能无关，而是与特征的局部性和粗粒度语义对齐有关。
摘要：
领先的视觉模型的可解释性如何？随着这些模型从研究基准转向高风险部署，这个问题变得日益紧迫，但现有方法无法可靠地回答这个问题。我们通过两种互补的心理物理学协议构建了一个框架来衡量和比较视觉模型的人类可解释性：（1）局部化性——观察者能否预测特征在新图像上的位置？（2）可命名性——观察者能否准确描述特征所代表的内容？特征通过稀疏自编码器恢复，一个基于偶然锚定的评分函数将每个模型置于同一尺度上。将该框架应用于六个视觉Transformer——两个监督ViT和四个基础模型（DINOv2、DINOv3、CLIP、SigLIP）——我们收集了超过15,000个行为响应，分析了377名通过我们预设质量检查的参与者中的13,400个响应。基础模型比其监督模型更难解释，且差距不是能力取舍：可解释性不与我们在任何基准上的下游任务性能相关。相关的是特征激活的局部性和与人类的粗粒度语义对齐——具有聚焦激活和反映世界广泛类别结构的表示模型产生更可解释的特征，而细粒度感知对齐则不。两种协议产生高度相关的排名并共享相同的预测因素，确立可解释性作为表示质量的独立、可测量的维度——令人惊讶的是，我们测试的每个基础模型都低于先前的监督基线。仅靠能力无法弥合这一差距；局部性和粗粒度对齐可以。
英文摘要：
How interpretable are the features of leading vision models? The question is increasingly pressing as these models move from research benchmarks into high-stakes deployments, yet existing methods cannot answer it reliably. We close this gap with a framework for measuring and comparing the human interpretability of vision models, built around two complementary psychophysics protocols: (1) localizability -- can an observer predict where a feature fires on a novel image? -- and (2) nameability -- can an observer accurately describe what the feature represents? Features are recovered via sparse autoencoders, and a chance-anchored scoring function places every model on a common scale. Applying the framework to six vision transformers -- two supervised ViTs and four foundation models (DINOv2, DINOv3, CLIP, SigLIP) -- we collected more than $15{,}000$ behavioral responses, analyzing the $13{,}400$ responses from the $377$ participants who passed our pre-specified quality checks. Foundation models are consistently *less* interpretable than their supervised counterparts, and the gap is not a capability tradeoff: interpretability does not correlate with downstream task performance on any benchmark we examine. What does correlate is the locality of a feature's activations and coarse-grained semantic alignment with humans -- models with focal activations and representations that reflect the world's broad categorical structure produce more interpretable features, whereas fine-grained perceptual alignment does not. The two protocols yield strongly correlated rankings and share the same predictors, establishing interpretability as an independent, measurable dimension of representation quality -- and, surprisingly, one on which every foundation model we tested falls below the supervised baselines that came before. Capability alone cannot close that gap; locality and coarse-grained alignment can.
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。
