English(EN) FARO: Function-Space Adaptive Return-to-Risk Optimization via Cross-Block Equal-Risk-Contribution Budget Allocation

新的FARO优化器通过重新分配风险预算来改进LLM训练

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-02 05:00

研究人员推出了一种新颖的优化框架FARO，旨在改进大型语言模型的训练。FARO将参数块视为基本单元，根据其反事实回报动态分配几何形状，从而调整优化过程。这种方法使用类似于投资组合理论的风险平价原则来重新分配更新预算。在GPT风格模型在FineWeb数据集上的实验中，FARO的自适应臂与现有手工设计方法的性能相当，而其ERC变体在多个种子上的损失减少方面取得了适度但持续的改进。 AI

影响这项新的优化技术可能导致更高效、更鲁棒的大型语言模型训练，从而可能降低计算成本并提高性能。

排序理由该集群包含一篇详细介绍用于训练大型语言模型的新优化方法的论文。[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: 1 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍用于训练大型语言模型的新优化方法的论文。[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: 1 days old; Coverage has settled into its steady-state source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Xiaojun Wu, Cehao Yang, Xuhui Jiang, Chengjin Xu, Jia Li, Jian Guo · 2026-11-02 05:00

FARO: Function-Space Adaptive Return-to-Risk Optimization via Cross-Block Equal-Risk-Contribution Budget Allocation

arXiv:2610.40858v1 Announce Type: cross Abstract: Training a large language model is compute-bound, and current recipes divide the work between optimizers by hand: orthogonalized momentum for attention and MLP matrices, Adam-style updates for the rest. This division rests on huma…

本文如何被排名

报道来源 [1]

FARO: Function-Space Adaptive Return-to-Risk Optimization via Cross-Block Equal-Risk-Contribution Budget Allocation

相关实体

相关话题