English(EN) Influence Is the Target-Response Jacobian Times Residuals

新的目标-响应雅可比矩阵在AI模型分析中优于影响度量

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-21 04:00

研究人员引入了目标-响应雅可比矩阵（$J$），这是一种用于理解模型预测如何响应单个训练目标变化的度量。该度量通过关注目标敏感性而非观测权重来补充现有的训练数据归因（TDA）方法。在包括tanh网络和MNIST分类器在内的各种数据集和网络架构上进行的实验表明，$J$在预测目标扰动方面显著优于传统的影响度量，在影响度量失败的情况下取得了接近完美的$R^2$分数。 AI

影响引入了一种分析模型行为的新度量，有望提高可解释性和调试能力。

排序理由学术论文，介绍新的技术概念和实验结果。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 0 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文，介绍新的技术概念和实验结果。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, other; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: 4 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.LG TIER_1 English(EN) · Paul Geertsema, Simon Geertsema · 2026-10-21 04:00

影响是目标-响应雅可比乘以残差

arXiv:2610.27689v1 Announce Type: new Abstract: We introduce the target-response Jacobian $J = [\partial \hat y_i/\partial y_j]$, which measures how model predictions respond to changes in individual training targets. $J$ complements training-data attribution (TDA) methods, which…

本文如何被排名

报道来源 [1]

影响是目标-响应雅可比乘以残差

相关实体

相关话题