arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2307.04719cs.LG
收藏

关于损失景观曲率的研究

On the curvature of the loss landscape

Technical University of Denmark(丹麦技术大学);ETH Zurich(苏黎世联邦理工学院)

Alison Pouplin, Hrittik Roy, Sidak Pal Singh, Georgios Arvanitidis

AI总结:

本文将深度网络的损失景观建模为黎曼流形,通过解析计算标量曲率,从微分几何角度建立其与模型泛化能力之间的联系。

摘要:

现代深度学习的主要挑战之一是理解为何这些过度参数化模型在有限数据上训练时表现如此出色。分析这种泛化概念的一种方法是通过相关损失景观的性质。在这项工作中,我们将损失景观视为一个嵌入的黎曼流形,并证明该流形的微分几何性质可用于分析深度网络的泛化能力。特别是,我们关注标量曲率,该曲率可以在我们的流形上解析计算,并展示了与若干可能蕴含泛化能力的设定之间的联系。

英文摘要:

One of the main challenges in modern deep learning is to understand why such over-parameterized models perform so well when trained on finite data. A way to analyze this generalization concept is through the properties of the associated loss landscape. In this work, we consider the loss landscape as an embedded Riemannian manifold and show that the differential geometric properties of the manifold can be used when analyzing the generalization abilities of a deep net. In particular, we focus on the scalar curvature, which can be computed analytically for our manifold, and show connections to several settings that potentially imply generalization.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑