leon.bottou.org

User Tools

Site Tools

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision 		Previous revision
papers:bordes-bottou-gallinari-2009 [2010/09/23 16:22]
leonb 		papers:bordes-bottou-gallinari-2009 [2017/11/29 10:27] (current)
leonb [Errata]
Line 12: Line 12:
 PASCAL Large Scale Learning Challenge. PASCAL Large Scale Learning Challenge.
  
-//Note//: +<html><font color=blue></html> 
-The appendix contains a derivation of upper and lower bounds  +//Errata//:  
-on the asymptotic convergence speed of stochastic gradient algorithm. +Please see section [[#Errata]] below. 
-This result is exact in the case of second order stochastic gradient. +<html></font></html>
  
 <box 99% orange> <box 99% orange>
Line 22: Line 21:
  
 [[http://jmlr.csail.mit.edu/papers/v10/bordes09a.html|JMLR Link]] [[http://jmlr.csail.mit.edu/papers/v10/bordes09a.html|JMLR Link]]
 +[[http://jmlr.csail.mit.edu/papers/v11/bordes10a.html|JMLR Erratum]]
 +<html>&nbsp;&nbsp;</html>
 [[http://leon.bottou.org/publications/djvu/jmlr-2009.djvu|jmlr-2009.djvu]] [[http://leon.bottou.org/publications/djvu/jmlr-2009.djvu|jmlr-2009.djvu]]
 [[http://leon.bottou.org/publications/pdf/jmlr-2009.pdf|jmlr-2009.pdf]] [[http://leon.bottou.org/publications/pdf/jmlr-2009.pdf|jmlr-2009.pdf]]
Line 48: Line 49:
  
  
-==== Erratum ====+==== Appendix ==== 
 + 
 +The appendix contains a derivation of upper and lower bounds  
 +on the asymptotic convergence speed of stochastic gradient algorithm. 
 +The constants are exact in the case of second order stochastic gradient. 
 + 
 + 
 +==== Errata ==== 
 + 
 +The SGDQN algorithm as described in this paper contains a subtle flaw 
 +described in a subsequent [[:papers:bordes-2010|erratum]]. 
 + 
 +There is a missing 1/2 factor in the bounds of theorem 1. 
 + 
 +\[ 
 + \def\w{\mathbf{w}} 
 + {\frac{1}{2}} \frac{{\mathrm tr}(\mathbf{HBGB})}{2\lambda_{\max}-1}\,t^{-1} + {\mathrm o}(t^{-1}) 
 +  ~\leq~ \mathbb{E}_{\sigma}\big[\:{\cal P}_n(\w_t)-{\cal P}_n(\w^*_n)\:\big] ~\leq~  
 +  {\frac{1}{2}} \frac{{\mathrm tr}(\mathbf{HBGB})}{2\lambda_{\min}-1}\,t^{-1} + {\mathrm o}(t^{-1})  
 +\] 
 + 
 +The version of the paper found on this site contains the correct theorem and proof. 
  
-There is an erratum for this paper:  
-[[http://jmlr.csail.mit.edu/papers/v11/bordes10a.html|JMLR, 11:2229−2240, 2010]], 
-{{jmlr_erratum_draft.pdf|pdf}}. 
  
  
  
  
papers/bordes-bottou-gallinari-2009.1285273378.txt.gz · Last modified: 2010/09/23 16:22 by leonb

Page Tools