Skip to content

eneftci/eCD

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

#-----------------------------------------------------------------------------
# File Name : README.py
# Purpose: README file
#
# Author: Emre Neftci
#
# Creation Date : 27-06-2014
# Last Modified : 
#
# Copyright : (c) UCSD, Emre Neftci, Srinjoy Das, Bruno Pedroni, Kenneth Kreutz-Delgado, Gert Cauwenberghs
# Licence : GPLv2
#----------------------------------------------------------------------------- 


Python pre-requisites
---------------------
-Brian (http://briansimulator.org/) version 1.4.0 WARNING: incompatible with version 1.4.3!
-Numpy (http://www.numpy.org/)
-Matplotlib (http://matplotlib.org/)

For kldivergence.py only:
-neuro_kl (https://github.com/pberkes/neuro-kl)


Other pre-requisites:
---------------------
- git (http://git-scm.com/)

Installation steps
------------------

>>> git clone  http://github.com/eneftci/eCD

Running the scripts
-------------------

The scripts in the experiments directory generate the data necessary for the figures in the paper "Event-driven Constrastive Divergence for Spiking Neuromorphic Systems".

calibrate.py: Calibrates the parameters of the sigmoid function by fitting the parameters gamma and beta to the transfer function of the IF neuron (Fig. 2)

accuracy_pre_trained.py
Tests recognition accuracy on the entire MNIST test-set 

accuracy_pre_trained_finite.py
Same as above but discretizes parameter to finite precision.

train_mnist.py
Main script for training the RBM using eCD

convergence.py
Generates (panels for Fig. 7) from a pre-trained RBM showing the convergence of the eCD rule. 

demonstrate_generative_model.py
Generates data for Fig. 9

The following files run the RBM + eCD (CD) on MNIST data, they can be ran directly, but are primarily intended to be used through the previously described scripts.

MNIST_IF_STDP.py: Script containing the main functions for training the Integrate and Fire neuron based RBM using event-driven CD
    MNIST_IF_STDP_SEQ.py: Same as above, but with modification for plotting purposes in demonstrate_generative_model.py
    MNIST_IF_STDP_WMON.py: Same as above, but with modification for plotting purposes in convergence.py
MNIST_SRM_STDP.py: Similar to above but using the ideal transfer curve (i.e. the Spike Response Model)
MNIST_SRM_RATE.py: Similar to above, but using standard CD and the ideal transfer curve (i.e. the Spike Response Model)

parameters_*.py: These files contain parameters used for the experiments scripts 


File permissions convention
---------------------------
Files that are designed to run standalone (with a if __name__ == '__main__' statement) have executable permissions, the others do not.


Remarks about customization
---------------------------
The neuron models are specified in neusa/experimentLib.py





About

Event-Driven CD in Spiking Neural Networks

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages