Plasticity-Mediated Competitive Learning

N. N. Schraudolph and T. J. Sejnowski. Plasticity-Mediated Competitive Learning. In Advances in Neural Information Processing Systems (NIPS), pp. 475–480, The MIT Press, Cambridge, MA, 1995.
In Ph.D. thesis

Download

pdf djvu ps.gz
183.0kB   76.3kB   48.4kB  

Abstract

Differentiation between the nodes of a competitive learning network is conventionally achieved through competition on the basis of neural activity. Simple inhibitory mechanisms are limited to sparse representations, while decorrelation and factorization schemes that support distributed representations are computationally unattractive. By letting neural plasticity mediate the competitive interaction instead, we obtain diffuse, nonadaptive alternatives for fully distributed representations. We use this technique to simplify and improve our binary information gain optimization algorithm for feature extraction (Schraudolph & Sejnowski, 1993); the same approach could be used to improve other learning algorithms.

BibTeX Entry

@inproceedings{SchSej95,
     author = {Nicol N. Schraudolph and Terrence J. Sejnowski},
      title = {\href{http://nic.schraudolph.org/pubs/SchSej95.pdf}{
               Plasticity-Mediated Competitive Learning}},
      pages = {475--480},
     editor = {Gerald Tesauro and David S. Touretzky and Todd K. Leen},
  booktitle =  nips,
  publisher = {The {MIT} Press, Cambridge, MA},
     volume =  7,
       year =  1995,
   b2h_type = {Top Conferences},
  b2h_topic = {>Competitive Learning},
   b2h_note = {In <a href="b2hd-Schraudolph95">Ph.D. thesis</a>},
   abstract = {
    Differentiation between the nodes of a competitive learning network is
    conventionally achieved through competition on the basis of neural
    activity.  Simple inhibitory mechanisms are limited to sparse
    representations, while decorrelation and factorization schemes that
    support distributed representations are computationally unattractive.
    By letting neural {\em plasticity}\/ mediate the competitive interaction
    instead, we obtain diffuse, nonadaptive alternatives for fully distributed
    representations.  We use this technique to simplify and improve our binary
    information gain optimization algorithm for feature extraction
    \href{b2hd-nips92}{(Schraudolph \& Sejnowski, 1993)}; the same
    approach could be used to improve other learning algorithms.
}}

Generated by bib2html.pl (written by Patrick Riley) on Thu Sep 25, 2014 12:00:33