SenseLab Home ModelDB Home

Robust Reservoir Generation by Correlation-Based Learning (Yamazaki & Tanaka 2008)
Accession: 116806
"Reservoir computing (RC) is a new framework for neural computation. A reservoir is usually a recurrent neural network with fixed random connections. In this article, we propose an RC model in which the connections in the reservoir are modifiable. ... We apply our RC model to trace eyeblink conditioning. The reservoir bridged the gap of an interstimulus interval between the conditioned and unconditioned stimuli, and a readout neuron was able to learn and express the timed conditioned response."
Reference: Yamazaki T, Tanaka S (2009) Robust Reservoir Generation by Correlation-Based Learning Advances in Artificial Neural Systems 2009:1-7
Citations  Citation Browser
Model Information (Click on a link to find other models with that property)
Model Type:  Network;
Brain Region(s)/Organism:  
Cell Type(s):   
Channel(s):   
Gap Junctions:  
Receptor(s):  
Gene(s):  
Transmitter(s):  
Simulation Environment:  C or C++ program;
Model Concept(s):  Temporal Pattern Generation; Spatio-temporal Activity Patterns; Rate-coding model neurons; Learning;
Implementer(s):  
Model files   Download zip file             Help downloading and running models
\
aans2008
README.html
button.png
Makefile
mt19937ar-cok.c
main.c
run.sh
w.0.bz2
xcorr.c
                            
* README

** What is this program?

This is the source code of my reservoir computing model presented in
the following paper:

    Tadashi Yamazaki and Shigeru Tanaka,
    Robust Reservoir Generation by Correlation-Based Learning,
    Advances in Artificial Neural Systems,
    vol. 2009, Article ID 467128, 7 pages, 2009. doi:10.1155/2009/467128

    http://www.hindawi.com/GetArticle.aspx?doi=10.1155/2009/467128

    This is an open access article so that anyone can download the PDF
    from the above link.

The model can be run online at the sim.neuinf.jp simulation platform: 


** Files

This folder contains the following files:

README :: This file
Makefile :: Makefile
main.c :: The simulation program
xcorr.c :: The program to calculate a similarity index
mt19937ar-cok.c :: The Mersenne Twister pseudo random number generator (*)
w.0.bz2 :: Initial connection weights (expand first by % bunzip2 w.0.bz2)

(*) http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html

** Usage

1. % make
2. % ./main 3 o w.0 w.1 33
3. % xcorr o.a o.a c

#1 generates main and xcorr binaries.

#2 generates o.a, o.r, and w.1.  File o.a is used to compute a similarity
index.  File o.r is used to obtain a raster plot.  For example, using
GNUPLOT, we do:

% gnuplot
gnuplot> plot [0:1000][0:100] 'o.r' w d

to plot the activity of the first 100 neurons for 1000 steps.  File w.1
is the updated connection weights from the initial weights specified by
w.0.  The 1st arg (3) is the random number seed for random connections
from excitatory to inhibitory neurons, so that this arg is fixed during
repeated trials.  The last arg (33) is the seed for temporal afferent
noise, so that this arg should change across trials.  See usage by just
% ./main.

#3 generates c.png, c.si, c.d, the similarity index, from file o.a.  
File c.png is the PNG file of the whole index matrix (Eq. 6).
File c.si contains the average and the standard deviation of the index
with the following format:

ModelDB Home  SenseLab Home   Help
Questions, comments, problems? Email the ModelDB Administrator
How to cite ModelDB
This site is Copyright 2012 Shepherd Lab, Yale University