English(EN) OrScale: Orthogonalised Optimization with Layer-Wise Trust-Ratio Scaling

OrScale优化方法增强神经网络训练

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-01 04:00

研究人员推出了一种新颖的优化方法OrScale，旨在改进大型神经网络的训练。OrScale通过将信任比率原则适配到正交化设置中，来解决更新的方向和幅度问题。该方法旨在提供动态的、每层的标量调整，确保优化收益有效地分配到不同的架构组件中。在从1.25亿到160亿参数的模型上进行的实验表明，OrScale在无需额外超参数调优的情况下，可以媲美甚至超越Muon等现有方法。 AI

影响引入了一种新颖的优化技术，可能导致更高效的大规模AI模型训练。

排序理由这是一篇详细介绍神经网络新优化方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 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: 23 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 这是一篇详细介绍神经网络新优化方法的学术论文。[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, infra; 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: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Yuxuan Lou, Yang You · 2026-09-01 04:00

OrScale：分层信任比率缩放的正交优化

arXiv:2605.07815v2 Announce Type: replace-cross Abstract: Muon fixes the \emph{direction} of every matrix-valued update at the polar factor of its momentum, while each layer's step \emph{magnitude} is addressed only by a static shape correction. We derive a dynamic per-layer scal…

本文如何被排名

报道来源 [1]

OrScale：分层信任比率缩放的正交优化

相关实体

相关话题