发表机构
Computer Vision Center; Universitat Autònoma de Barcelona; Warsaw University of Technology; IDEAS Research Institute（计算机视觉中心; 巴塞罗那自治大学; 华沙理工大学; IDEAS研究所）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对文本到图像模型中概念过渡的非线性特性，提出KANSteer方法，利用Kolmogorov-Arnold网络建模曲线激活轨迹，实现比线性引导更平滑的中间属性遍历。
AI 中文摘要
机制可解释性通常依赖于线性表示假设（LRH），该假设认为高层概念在激活空间中被编码为线性方向。然而，自然的视觉概念并不一定需要线性的视觉过渡：在晴朗和暴风雨之间存在着一个中间天气状态，例如带有几朵白云的天空，而不仅仅是较弱的暴风雨；在毛毛虫和蝴蝶之间，其进展也不是一只翅膀不断生长的毛毛虫。这引发了一个问题：这种真正的中间状态是否也被模型非线性地表示。确实，当我们直接提示文本到图像模型生成中间属性时，其激活很少落在连接端点的直线方向上。因此，我们提出了KANSteer，将概念遍历建模为一条穿过其中间状态的曲线。为了寻求既简单又可解释的表示，我们建议使用Kolmogorov-Arnold网络（KANs），它提供一个一维坐标，其学习到的函数定义了轨迹。这使得引导方向在概念上可以变化，同时保持可解释的表示。在多个概念和文本到图像扩散Transformer上，我们发现它们的激活轨迹显著偏离直线，并且KANSteer比线性引导提供了更紧密的拟合和更平滑的中间属性遍历。
英文摘要
Mechanistic interpretability often relies on the Linear Representation Hypothesis (LRH), which assumes that high-level concepts are encoded as linear directions in activation space. Yet a natural visual concept does not necessarily require a linear visual transition: between sunny and stormy lies an intermediate weather state such as a sky with a few white clouds, not simply a weaker storm; between a caterpillar and a butterfly, the progression is not a caterpillar with continuously growing wings. This raises the question of whether such true intermediate states are also represented nonlinearly by the model. Indeed, when we prompt text-to-image models directly for intermediate attributes, their activations rarely fall along the straight direction connecting the endpoints. Therefore, we propose KANSteer, which models concept traversal as a curve passing through its intermediate states. Seeking a representation that is both simple and interpretable, we propose to use Kolmogorov-Arnold Networks (KANs), which provide a one-dimensional coordinate whose learned functions define the trajectory. This allows the steering direction to vary along the concept while preserving an interpretable representation. Across several concepts and text-to-image diffusion transformers, we find that their activation trajectories substantially deviate from straight lines, and that KANSteer provide a closer fit and smoother traversal of intermediate attributes than linear steering.
