PulseAugur
中
实时 00:00:53
English(EN) The Road Taken: The Role of Optimizers at the Edge of Stability

新理论重塑深度学习中的优化器稳定性

研究人员在深度学习中发现了一种现象,即基于梯度的优化器在理论预测的不稳定阈值之上保持稳定的Hessian特征值。这种偏差,观察到高达预测边界的21.1倍,是系统性的,并且取决于所使用的特定优化器。该研究提出了一种基于方向Hessian和梯度对齐得分的稳定性阈值的新表述,该表述考虑了优化器的实际更新,并提供了新的诊断工具来理解其在优化过程中平衡时间和空间预算中的作用。 AI

影响 深化对优化动力学的理解,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,详细阐述了深度学习中优化器稳定性的新理论表述。[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 →

新理论重塑深度学习中的优化器稳定性

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,详细阐述了深度学习中优化器稳定性的新理论表述。[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
49 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]

  1. arXiv cs.LG TIER_1 English(EN) · Jaerin Lee, Kyoung Mu Lee ·

    所选之路:优化器在稳定性边缘的作用

    arXiv:2608.18415v1 Announce Type: new Abstract: The edge of stability refers to a phenomenon in deep learning with gradient-based optimizers where the Hessian eigenvalues of the loss remain stable above a threshold that the classical descent lemma predicts to be unstable. Previou…