Aufgabe 7: Liquid State Machines

[10+3* Punkte, ausgegeben am 24.05.2005, Abgabe bis 07.06.2005, pdf, ps.gz]

In this exercise you have to perform some experiments which will help you in understanding the concept of liquid state machine (LSM). For this exercise you have to use the lsm.tar.gz (download tar.gz, download tar.zip) which contains all the source code that you will need to perform your experiments. You can also download the documents csim-1.0-usermanual.pdf, circuit-tool-1.0-manual.pdf, and learning-tool-1.0-manual.pdf,from http://www.lsm.tugraz.at which provide a tutorial on using this code. This code is guaranteed to work only on linux machines and with matlab 6.5 (there might be problems with matlab 7). So it is recommended that you work on these exercises in our student lab.

1 Description

In this exercise you have the following setup: You have a recurrent spiking neural network, also called as the liquid, that is composed of 135 leaky-integrate-and-fire neurons. This liquid is receiving 2 timevarying poisson spike trains as the input. Three linear readouts are connected to all the neurons in this liquid. These linear readouts are supposed to calculate three different target functions. The first readout has to find the sum of rates of the two input spike trains in the last 30 msec. The second readout has to find the spike correlations in the input channels in the last 75 msec. The third readout should calculate at time $ t$ the sum of rate of the input channels in the time window $ [t-50 t-20]$ msec.

2 Setting up the System

  1. First, unzip the lsm.tar.gz tarball into your home directory. Now change your working directory to HOME/lsm/learning/demos/.
  2. You will see 4 different directories called ex_lsm , multi , segment_classification, and spike_train_classification here. The last 3 are fully working demos which have been supplied to you so that you can become comfortable with playing around with the system. The ex_lsm directory is where you shall be working for this exercise.
  3. In each of the demo directories there is a startup-file startup.m. If this file is not executed automatically in matlab, you should run it once at the start of your matlab session to initialize the paths.
  4. To play around with the demos, just change your working directory to one of the directoreis, e.g. segment_classification . Run matlab and on the command prompt type seg_class. The simulation should run smoothly.
  5. Now go back to the ex_lsm directory.

3 Experiments

Run the code by typing ex_lsm in matlab in this directory.

You can change circuit parameters in the file make_liquid.

  1. Vary the recurrency inside the liquid (The parameter lambda changes the ammount of recurrent connection, whereas the parameter Wscale scales the strength of connections). Describe your observations about the effects of recurrency on the performance (collect some statistics and emphasize your results with some plots). Is recurrency helpful here?
  2. Vary the size of the training set. Describe differences in performance.
  3. Try to find an ideal setting of parameters (you can also change connectivity parameters from the inputs to the liquid). What performance can you achieve?

Neuronale Netzwerke B (SS05), Rober Legenstein, 2005