Papers

arXiv 论文提出解码器感知归因方法,修正结构化输出模型的输入解释偏差

Attribution Through the Decoder: Prediction-Space Explanations for Structured Outputs

Why we picked it

做姿态估计、视线追踪这类方向预测的工程师可以看看这篇:它证明直接解释原始坐标会把注意力分给解码器丢掉的分量,合成任务上高达 72% 归因质量落在无关区域。换成解码器微分目标后,MPIIGaze 上删除像素的位移提升 44%,比普通 Grad-CAM 解释更可信。

一篇 arXiv 论文指出,预测角度、视线方向、旋转等结构化输出的网络通常先输出原始坐标,再由归一化等固定解码器映射为最终预测,而现有归因方法直接作用于原始坐标,会把权重分给解码器丢弃的分量(如不影响方向的向量缩放)。作者提出直接穿过解码器微分做归因,对归一化解码器等价于缩放后的切向投影,并刻画了输出缩放不变性与球面标量目标的零切向敏感点性质。

Body

实验给出具体数字:在解析相关已知的合成任务上,环境空间归因最多把 72% 的质量分给无关区域;在 MPIIGaze 验证集上,切向 Grad-CAM 在 10% 像素删除下产生 44% 更大的预测位移;在 PASCAL3D+ 上配合 6D 旋转解码器,完整微分方法的平均删除位移比环境坐标求和高 14.1 度,比解码器不变的标量化高 13.8 度。两极性切向 LayerCAM 结果不一,圆形回归上优于环境求和版本,四元数姿态上则无显著差异。

论文将失败来源拆分为径向污染与标量化损失两类,并提醒不变归因仍可能漏掉真正驱动预测的输入,需同时检查不变性与删除性能。该方法复用现有归因技术,只需换成解码器感知的目标,部分变体需改动归因聚合方式。

Original · arXiv cs.AI

Attribution Through the Decoder: Prediction-Space Explanations for Structured Outputs

Networks that predict angles, gaze directions, and rotations do not output these quantities directly: they emit raw coordinates that a fixed decoder, such as normalization, maps to the final prediction. Attribution methods are usually applied to the raw coordinates, so they can credit inputs for changes the decoder discards, such as rescaling a vector when only its direction matters. The resulting explanation can highlight inputs that have no effect on the prediction. We instead attribute through the decoder differential, which for normalization decoders reduces to a scaled tangent projection. We characterize invariance to output rescaling, show that every fixed smooth scalar target on a sphere has points of zero tangential sensitivity, and show that an endpoint-projected variant of Integrated Gradients can still attribute to scale-only inputs, whereas direct angular Integrated Gradients cannot. On synthetic tasks with analytically known relevance, ambient attribution assigns up to 72% of its mass to an irrelevant region. On MPIIGaze validation data, tangent Grad-CAM produces 44% greater prediction displacement under 10% pixel deletion. On PASCAL3D+ with a 6D rotation decoder, the full differential improves mean deletion displacement by 14.1 degrees over the ambient coordinate sum and by 13.8 degrees over a decoder-invariant scalar. Results for a two-polarity tangent LayerCAM variant are mixed: it exceeds ambient-sum LayerCAM on circular regression but no difference is detected on quaternion pose. These results separate two failure sources, radial contamination and scalarization loss. The approach uses existing attribution methods with decoder-aware targets, although some variants also require changes to attribution aggregation. An invariant attribution can still miss inputs that drive the prediction, so invariance and deletion performance should each be checked.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Integrated Gradients, Grad-CAM, interpretability, papers.

Source attribution: curated from arXiv cs.AI (@Zaer Bin Zaqy, Brinnae Bent).