Statistics > Machine Learning
[Submitted on 1 Mar 2018 (v1), revised 7 Mar 2018 (this version, v3), latest version 10 Jun 2019 (v5)]
Title:The Regularization Effects of Anisotropic Noise in Stochastic Gradient Descent
View PDFAbstract:Understanding the generalization of deep learning has raised lots of concerns recently, where the learning algorithms play an important role in generalization performance, such as stochastic gradient descent (SGD). Along this line, we particularly study the anisotropic noise introduced by SGD, and investigate its importance for the generalization in deep neural networks. Through a thorough empirical analysis, it is shown that the anisotropic diffusion of SGD tends to follow the curvature information of the loss landscape, and thus is beneficial for escaping from sharp and poor minima effectively, towards more stable and flat minima. We verify our understanding through comparing this anisotropic diffusion with full gradient descent plus isotropic diffusion (i.e. Langevin dynamics) and other types of position-dependent noise.
Submission history
From: Zhanxing Zhu [view email][v1] Thu, 1 Mar 2018 03:46:36 UTC (828 KB)
[v2] Fri, 2 Mar 2018 10:19:15 UTC (827 KB)
[v3] Wed, 7 Mar 2018 08:16:16 UTC (827 KB)
[v4] Mon, 21 May 2018 15:01:58 UTC (873 KB)
[v5] Mon, 10 Jun 2019 05:19:32 UTC (8,783 KB)
Current browse context:
stat.ML
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.