English(EN) The Multiple Timescales of Gradient Descent on the Edge of Stability: A Perturbative Derivation of the Central Flow

新理论解释了稳定性边缘的梯度下降动力学

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-02 04:00

研究人员开发了一种新的扰动方法，用于在深度学习的稳定性边缘正式推导出梯度下降的中心流模型。该方法将梯度下降视为一个奇异摄动的动力学系统，揭示了三个不同的时间尺度：快速振荡、中间自稳定和沿最小化器的慢动力学。中心流作为该展开的领先阶项出现，而自稳定机制出现在后续项中，从而提供了对涨落持续性的更深入理解。 AI

影响为理解梯度下降提供了更严格的理论基础，有可能导致更稳定、更高效的深度学习模型训练。

排序理由学术论文，详细介绍了机器学习概念的新理论推导。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv stat.ML 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: 80 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文，详细介绍了机器学习概念的新理论推导。[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: 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 stat.ML TIER_1 English(EN) · Rapha\"el Berthier · 2026-09-02 04:00

梯度下降在稳定性边缘的多时间尺度：中心流的扰动推导

arXiv:2609.01034v1 Announce Type: cross Abstract: The central flow of Cohen et al. (2025) is an empirically accurate continuous-time model of gradient descent at the edge of stability in deep learning, However, its derivation is heuristic. We propose a perturbative regime in whic…

本文如何被排名

报道来源 [1]

梯度下降在稳定性边缘的多时间尺度：中心流的扰动推导

相关实体

相关话题