MIT Libraries logoDSpace@MIT

MIT
View Item 
  • DSpace@MIT Home
  • Center for Brains, Minds & Machines
  • Publications
  • CBMM Memo Series
  • View Item
  • DSpace@MIT Home
  • Center for Brains, Minds & Machines
  • Publications
  • CBMM Memo Series
  • View Item
JavaScript is disabled for your browser. Some features of this site may not work without it.

Loss landscape: SGD can have a better view than GD

Author(s)
Poggio, Tomaso; Cooper, Yaim
Thumbnail
DownloadCBMM-Memo-107.pdf (1.033Mb)
Metadata
Show full item record
Abstract
Consider a loss function L = 􏰀ni=1 l2i with li = f(xi) − yi, where f(x) is a deep feedforward network with R layers, no bias terms and scalar output. Assume the network is overparametrized that is, d >> n, where d is the number of parameters and n is the number of data points. The networks are assumed to interpolate the training data (e.g. the minimum of L is zero). If GD converges, it will converge to a critical point of L, namely a solution of 􏰀ni=1 li∇li = 0. There are two kinds of critical points - those for which each term of the above sum vanishes individually, and those for which the expression only vanishes when all the terms are summed. The main claim in this note is that while GD can converge to both types of critical points, SGD can only converge to the first kind, which include all global minima.
Date issued
2020-07-01
URI
https://hdl.handle.net/1721.1/126041
Publisher
Center for Brains, Minds and Machines (CBMM)
Series/Report no.
CBMM Memo;107

Collections
  • CBMM Memo Series

Browse

All of DSpaceCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

My Account

Login

Statistics

OA StatisticsStatistics by CountryStatistics by Department
MIT Libraries
PrivacyPermissionsAccessibilityContact us
MIT
Content created by the MIT Libraries, CC BY-NC unless otherwise noted. Notify us about copyright concerns.