English(EN) Gaussian Equivalence for Multi-Head Self-Attention

新理论表明高斯噪声可保留多头自注意力机制中的谱律

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-08 04:00

研究人员为多头自注意力机制建立了一个名为高斯等价性的理论框架。该框架表明，用重缩放分数加上高斯噪声替换标准的softmax注意力，可以保持中心化输出的极限谱律。这些发现利用随机矩阵理论得出，也扩展到依赖于键（keys）的值（values）和输出投影，区分了头分配和投影宽度对谱律保持的影响。 AI

影响提供了对现代神经网络核心组件的更深层次的理论理解，可能指导未来的架构设计。

排序理由阐述神经网络机制理论进展的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

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: 7 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 阐述神经网络机制理论进展的学术论文。[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: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

arXiv cs.LG TIER_1 English(EN) · Tomohiro Hayase, Ryo Karakida · 2026-10-08 04:00

多头自注意力机制的高斯等价性

arXiv:2610.10033v1 Announce Type: cross Abstract: A theoretical understanding of multi-head self-attention is fundamental to the study of modern neural networks. Using random matrix theory, we establish Gaussian equivalence for multi-head self-attention: replacing softmax attenti…