Skip to content

mudigonda/Minimum-Probability-Flow-Learning

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

35 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Minimum Probability Flow learning (MPF)

MPF is a technique for parameter estimation in un-normalized probabilistic models. It is described in the paper:

J Sohl-Dickstein, P Battaglino, MR DeWeese
Minimum probability flow learning
International Conference on Machine Learning (2011)
http://arxiv.org/abs/0906.4779

This repository contains Matlab code implementing MPF for the Ising model and the RBM. The directory structure is as follows:

  • MPF_ising/ - parameter estimation in the Ising model
  • MPF_RBM_compare_log_likelihood/ - parameter estimation in Restricted Boltzmann Machines. This directory also includes code comparing the log likelihood of small RBMs trained via pseudolikelihood and Contrastive Divergence to ones trained via MPF.

If you're interesting in using MPF to build an Ising model of neural spike data, you should also check out Liberty Hamilton's repository at https://github.com/libertyh/ising-model.

About

Matlab code implementing Minimum Probability Flow Learning.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • MATLAB 81.0%
  • C 11.8%
  • TeX 6.6%
  • Other 0.6%