LeGrad：一种通过特征形成敏感性实现 Vision Transformer 可解释性的方法
LeGrad: An Explainability Method for Vision Transformers via Feature Formation Sensitivity
AI总结：
本文提出面向 ViT 的可解释性方法 LeGrad，利用各层注意力图梯度并聚合多层 token 激活生成解释图，在分割、扰动和开放词表任务中展现出更优的空间保真度与鲁棒性。
摘要：
Vision Transformer（ViT）能够通过自注意力机制建模长程依赖，现已成为计算机视觉中的标准架构。然而，这些模型的可解释性仍然是一项挑战。为解决该问题，我们提出 LeGrad，这是一种专门为 ViT 设计的可解释性方法。LeGrad 计算相对于 ViT 各层注意力图的梯度，并将梯度本身视为可解释性信号。我们在所有层上聚合该信号，同时结合最后层以及中间 token 的激活，以生成合并后的可解释性图。这使 LeGrad 成为一个概念简单且易于实现的工具，可用于提升 ViT 的透明度。我们在具有挑战性的分割、扰动和开放词表设置中评估 LeGrad，展示了其相较于其他 SotA 可解释性方法的多功能性，并证明其具有更优的空间保真度和对扰动的鲁棒性。演示和代码可在 https://github.com/WalBouss/LeGrad 获取。
英文摘要：
Vision Transformers (ViTs), with their ability to model long-range dependencies through self-attention mechanisms, have become a standard architecture in computer vision. However, the interpretability of these models remains a challenge. To address this, we propose LeGrad, an explainability method specifically designed for ViTs. LeGrad computes the gradient with respect to the attention maps of ViT layers, considering the gradient itself as the explainability signal. We aggregate the signal over all layers, combining the activations of the last as well as intermediate tokens to produce the merged explainability map. This makes LeGrad a conceptually simple and an easy-to-implement tool for enhancing the transparency of ViTs. We evaluate LeGrad in challenging segmentation, perturbation, and open-vocabulary settings, showcasing its versatility compared to other SotA explainability methods demonstrating its superior spatial fidelity and robustness to perturbations. A demo and the code is available at https://github.com/WalBouss/LeGrad.
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。
