A mean field theory learning algorithm for neural networks
(1987) In Complex Systems 1. p.995-1019- Abstract
- Based on t he Boltzmann Machine concept, we derive a
lear ning algorithm in which time-consuming stochastic measurements
of correlations a re replaced by solutions to dete rminist ic mean field
theory equ ations. T he method is applied to t he XOR (exclusive-or ),
encoder, and line sym metry problems with substantial success. We
observe speedup facto rs ranging from 10 to 30 for these ap plicat ions
and a significan tly bet ter learning performan ce in general.
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/624aca74-da8b-41ae-a50b-906635925902
- author
- Peterson, Carsten LU and Anderson, James R
- organization
- publishing date
- 1987
- type
- Contribution to journal
- publication status
- published
- subject
- in
- Complex Systems
- volume
- 1
- pages
- 24 pages
- language
- English
- LU publication?
- yes
- id
- 624aca74-da8b-41ae-a50b-906635925902
- date added to LUP
- 2024-12-11 09:04:03
- date last changed
- 2025-04-04 15:14:33
@article{624aca74-da8b-41ae-a50b-906635925902, abstract = {{Based on t he Boltzmann Machine concept, we derive a<br/>lear ning algorithm in which time-consuming stochastic measurements<br/>of correlations a re replaced by solutions to dete rminist ic mean field<br/>theory equ ations. T he method is applied to t he XOR (exclusive-or ),<br/>encoder, and line sym metry problems with substantial success. We<br/>observe speedup facto rs ranging from 10 to 30 for these ap plicat ions<br/>and a significan tly bet ter learning performan ce in general.}}, author = {{Peterson, Carsten and Anderson, James R}}, language = {{eng}}, pages = {{995--1019}}, series = {{Complex Systems}}, title = {{A mean field theory learning algorithm for neural networks}}, volume = {{1}}, year = {{1987}}, }