Variable Metric Stochastic Approximation Theory

P. Sunehag, J. Trumpf, S. Vishwanathan, and N. N. Schraudolph. Variable Metric Stochastic Approximation Theory. In Proc. 12^th Intl. Conf. Artificial Intelligence and Statistics (AIstats), pp. 560–566, Journal of Machine Learning Research, Clearwater Beach, Florida, 2009.

Download

167.0kB	80.5kB	182.6kB

Abstract

We provide a variable metric stochastic approximation theory. In doing so, we provide a convergence theory for a large class of online variable metric methods including the recently introduced online versions of the BFGS algorithm and its limited-memory LBFGS variant. We also discuss the implications of our results for learning from expert advice.

BibTeX Entry

@inproceedings{SunTruVisSch09,
     author = {Peter Sunehag and Jochen Trumpf and
               S.~V.~N. Vishwanathan and Nicol N. Schraudolph},
      title = {\href{http://nic.schraudolph.org/pubs/SunTruVisSch09.pdf}{
               Variable Metric Stochastic Approximation Theory}},
     editor = {David {van Dyk} and Max Welling},
      pages = {560--566},
  booktitle = {Proc.\ 12$^{th}$ Intl.\ Conf.\ Artificial
               Intelligence and Statistics (AIstats)},
    address = {Clearwater Beach, Florida},
     volume =  5,
     series = {Workshop and Conference Proceedings},
  publisher =  jmlr,
       year =  2009,
   b2h_type = {Top Conferences},
  b2h_topic = {>Quasi-Newton Methods},
   abstract = {
    We provide a variable metric stochastic approximation theory.
    In doing so, we provide a convergence theory for a large class
    of online variable metric methods including the recently
    introduced online versions of the BFGS algorithm and its
    limited-memory LBFGS variant. We also discuss the implications
    of our results for learning from expert advice.
}}