Circuit-Tool : A Tool for generating Neural MicroCircuits User Manual
Contents
1. Again one can specify in addition to the destination and the source other parameters which determine how individual synaptic connections are created For example the parameter lambda determines the average distance of synaptic connections and SH_W determines the standard deviation SD of the Gamma distribution used to generate the synaptic weights SD SHW mean Figure shows a typical postsynaptic connectivity pattern for one neuron generated by the code above 2 3 3 Connecting pools In this example we connect the pool of HHneurons to the LifNeuron pool by static synapses gt gt nmc c 6 add nmc conn dest p_hh src p_lif gt SH_W 0 5 lambda Inf type StaticSpikingSynapse Note that using lambda Inf has the effect that the distance between the neurons dose not matter in determining wheter a synaptic connection will by generated or not Figure 5 Postsynaptic connections of a typical excitatory LifNeuron 2 4 Simulating the model Now that the model is set up we turn to the issue how to define the input and simulate the network with these inputs 2 4 1 Setting up the input As we have three input neurons 2 spiking one analog we have to define a stimulus which consists of three channels 2 spiking one analog The stimulus can be defined as follows Nn Nn channel 2 data channel 2 spiking 1 create empty structure empty_stimulus nCh2. Now we want to set up synaptic connections between the neurons in the individual pools This is done by commands of the form gt gt nmc c_idx add nmc conn dest lt destination gt src lt source gt ge i ete a y meme o 39 Oe e To re D t 3 ger e Ty e a oe e S 9S a 2 am eee wens Figure 3 Synaptic connections of the spiking input neurons to the LifNeuron pool where lt destination gt and lt source gt specify a set of neurons either by the handle index of a pool or by specifying a volume 2 3 1 Connecting the input To see how it works lets start by connecting the spiking inputs to the pool of LifNeurons gt gt nmc c 1 add nmc conn dest p_lif src p_sin gt Cscale 1 Wscale 5 Here we used the pool handles indices as source and destination specification The additional parameters given specify how to scale the overall connection probability Cscale and the synaptic strength Wscale To see the actual connectivity pattern generated you can again use the command plot nmc and interaktively explore the network structure By clicking on a neuron you can look at presynaptic as well as postsynaptic connections Figure shows how the input neurons are connected at the moment As the next step we connect the spiking input via static synapses Sec default are dynamic synapses Sec to some subset of th
3. 4 Simulating the model sss seses eden eee eaw ad khad aada eo 8 3 Stimulus and Response 10 3 1 The Stimuls or Input Signals 22566 es0K8 e454 ded e KO EG ww SOS 10 3 2 The response or the network output 0 2 0 02 eee eee 10 4 Input Distributions 11 5 Distributed simulations 11 6 neural_microcircuit class reference 11 7 delay_lines class reference 11 8 small_circuits class reference 11 1 Preliminaries 1 1 What is Circuit Tool Circuit Tool is a set of Matlab objects and scripts that allow the construction of multi column neural microcircuits with a distribution of parameters that match those reported in the literature This neural microcircuit models can then be simulated efficiently using CSIM 1 2 About this Manual This manual is intended to describe how to use Circuit Tool from the Matlab users point of view It does not try to explain or give an introduction to the type of models which can be constructed an simulated with Circuit Tool Regarding neural modeling we refer the reader to Dayan and Abbott 2001 and Gerstner and Kistler 2002 Furthermore Matlab programming knowledge is assumed 1 3 Features of the current version Customizeable intra and inter column connectivity Runs under Unix Linux and Windows Object oriented design Parallel simulation for large set of stimuli 1 4 Getting and Installing Circuit Tool Circuit Tool is distributed under the GNU General Public License
4. Circuit Tool A Tool for generating Neural MicroCircuits Version 1 0 User Manual c 2002 The IGI LSM Group http www 1lsm tugraz at June 11 2006 This document is part of Circuit Tool Release 1 0 Copyright 2002 The IGI LSM group Circuit Tool is free software you can redistribute it and or modify it under the terms of the GNU General Public License as published by the Free Software Foundation either version 2 or at your option any later version Circuit Tool is distributed in the hope that it will be useful but WITHOUT ANY WARRANTY without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE See the GNU General Public License for more details To get a copy of the GNU General Public License point your browser to http www gnu org copyleft gpl html The IGI LSM group Institute for Theoretical Computer Science Graz University of Technology Inffeldgasse 16 b A 8010 Graz AUSTRIA mailto lsm igi tu graz ac at http www 1lsm tugraz at Contents 1 Preliminaries 2 LI I CUO eass Soe ee eet ee Se ow we 2 12 About this Manual gt eses sesers raais ORE OES EO 2 1 3 Features of the current version o oo osoa e a e 3 1 4 Getting and Installing Circuit Tool oaa 2 0 0 000000 3 2 A short Tutorial 3 21 Initializing the model lt o ss sos sam omom OD a ee ee ce d 4 2A Creatine tbe Pools so s sss sasaaa eee ee ki soka a ew a a 4 2 3 Making synaptic connections aoao ee ee 5 2
5. and can be downloaded from http www igi tugraz at circuits To install Circuit Tool perform the following steps 1 Donwload Circuit Tool from www igi tugraz at circuits 2 Unzip the file circuits VER zip where VER stands for the version you have downloaded This will create a subdirectory 1sm and 1lsm circuits 3 Start Matlab and change into the directory 1sm 4 Run the Matlab script install m 5 Add the path 1sm to the Matlab search path e g e addpath home jack 1sm or e addpath C Work Neuroscience lsm 6 Change into the directory 1sm circuits demos and play around with them 7 Have fun using Circuit Tool 2 A short Tutorial In this section we will introduce Circuit Tool by means of an example We will use Circuit Tool to construct a model which consists of three columns or pools how they are called in Circuit Tool The first pool consists of leaky integrate an fire neurons the second is made up of sigmoidal neurons and the third uses more detailed conductance based model neurons The pools will be connected internally and to each other for details see below The circuit will be driven by two input spike trains and an analog input current The input projects in some kind of topographic map into the circuit Figure 1 A 3 x 3 x 10 pool of LifNeurons with origin 2 1 1 2 1 Initializing the model The Circuit Tool implementation of this model employs the Matlab class neural _microcircu
6. annels 3 Tstim 1 fill channel 1 with some spikes channel 1 data 1 rand 1 10 channel 1 spiking 1 fill channel 2 with some spikes 1 rand 1 20 channel 3 is a sine wave channel 3 dt 0 005 channel 3 data 1 sin 2 pi 10 0 S channel 3 dt 1 channel 3 spiking 0 Note that each signal channel can be either spiking S channel i spiking 1 or analog S channel i spiking 0 In the later case one has to specify the temporal resolution S channel i dt of the signal stimulus N o T channel 1 oe ey li l li ll i pamel 4 li A 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 time sec Figure 6 A stimulus with 2 spiking and one ananlog channel We adopt the convention that the first channel of the stimulus is assigned to the first input neuron created by means of nmc p add nmc pool statements Keep this convention in mind when setting up the stimulus otherwise you may be surprised by error messages or strange results if the input neurons have different connections to the rest of the network You can use the command gt gt plot_stimulus S to plot the stimulus defined above This results in the plot shown in Figure 2 4 2 Defining the respones Obviously we want to see how the network responses to the given stimulus Therefor we must specify what we want to record during the stimulation The following code fragment shows how to record the spikes
7. e pool of sigmoidal neurons by specifying a certain volume gt gt mmc c 2 add nmc conn dest 6 1 1 6 3 6 src p_sin type StaticSpikingSynapse Cscale Inf The synaptic connection created by this command are shown in Figure Note that the setting Cscale Inf ensures that there will be a synaptic connection between each pair of neurons in the source region and the destination region And finaly we connect the analog input to the pool of HHNeurons by means of a StaticAnalogSynapse This is necessary since StaticSpikingSynapses can not transmit analog signals gt gt nmc c 3 add nmc conn dest p_hh src p_ain type StaticAnalogSynapse Wscale 0 2 Figure 4 Synaptic connections of the spiking input neurons to the SigmoidalNeuron pool 2 3 2 Making recurrent connections Now we want to create recurrent connections with the pools themselves The only differene to the previous section is that now the source and destination is the same pool gt gt nmc c 4 add nmc conn dest p_lif src p_lif gt SH_W 0 5 lambda 2 5 Wscale 2 gt gt nmc c 5 add nmc conn dest p_sig src p_sig gt SH_W 0 5 lambda 2 type StaticAnalogSynapse gt gt nmc c 7 add nmc conn dest p_hh src p_hh gt SH_W 0 5 lambda 1
8. es or as an analog signal e S channel i dt time discretization for analog signals S channel i spiking 0 only e S channel i data signal date vektor of the analog values S channel i spiking 0 or spike times S channel i spiking 1 e S info i Tstim the length of the stimulus usally used in plotting routines 3 2 The response or the network output After a network simulation via a command like gt gt R simulate nmc Tstim S 10 the response is stored in the cell array R R i contains the traces specified by the i th nmc record nmc statement during the setup of the simulation R i by itself is a struct array with the only field channel which is in turn a struct array with a similar structure as an input signal Sec 3 1 That is e R i channel j data signal data vektor of the analog values or spike times Note that the data always starts at time t 0 e R i channel j spiking binary flag 0 1 which determines if data should be interpreted as spike times or as an analog signal e R i channel j dt time discretization for analog signals only e R i channel j fieldName name of the recorded field e R i channel j idx handle of the object from which field the data was recorded Aa Input Distributions 5 Distributed simulations 6 neural_microcircuit class reference J delay_lines class reference 8 small_circuits class reference References Dayan and Abbott 2001 Da
9. ifNeuron analog input neuron Figure 2 Three pools of model neurons consisting of neurons of the classes LifNeuron Sig moidalNeuron and HHNeuron from left to right and two input pools size 3 3 6 origin 6 1 1 Neuron thresh 1 Neuron beta 2 Neuron tau_m 3 Neuron A_max 4 Neuron I_inject 1 Neuron Vm_init 0 gt gt nmc p_hh add nmc pool type HHNeuron size 3 3 6 origin 10 1 1 Neuron Inoise Neuron Iinject The code fragment above shows how to set off default values for some parameters of the Neurons generated One has to add a pair of Neuron lt field gt lt value gt arguments to the function call Which fields are valid is determined by the class of the neuron see the CSIM Class Reference for details As the next step we create the input neurons A pool of 2 excitatory spiking input neurons gt gt nmc p_sin add nmc pool type SpikingInputNeuron size 1 1 2 origin O 1 5 frac_EXC 1 0 and a pool of a single excitatory analog input neuron gt gt nmc p_ain add nmc pool type AnalogInputNeuron gt size 1 1 1 origin O 1 2 frac_EXC 1 0 A visualization of the current model is shown in Figure which was produce by the command plot nmc 2 3 Making synaptic connections
10. it Sec 6 To start our construction of the model we will instantiate an empty microcircuit gt gt mmc neural_microcircuit 2 2 Creating the Pools As the next step we will create the individual pools First we will create a pool of integrate and fire neurons gt gt nmc p_lif add nmc pool type LifNeuron size 3 3 6 origin 2 1 1 The above command creates 54 neurons of the class LifNeuron which is a neuron type available in CSIM and adds them to nmc Note that due to the object oriented paradigm used in Matlab the nmc object must appear also on the left hand side of the command The variable p_lif is a handle index to refer to that particular pool later in the programm The neurons are located on a three dimensional 3 x 3 x 6 integer grid with origin 2 1 1 You can visualize this by issuing the command gt gt plot nmc The plot command should produce a figure which looks very much like Figure As you can see in Figure some neurons are marked by magenta balls These are inhibitory neurons while the other are excitatory neurons By default a neuron is choosen to be a excitatory with a probability of 80 this can be controlled with the frac_EXC parameter Now we add the other two pools of neurons where we set some off default parameters gt gt nmc p_sig add nmc pool type SigmoidalNeuron HHNeuron spiking input neurons SigmoidalNeiron L
11. of the pool of LifNeurons and the membrane voltage of a certain subset defined by specifiyng the appropriate volume of the SigmoidalNeuron and HHNeuron pool gt gt nmc record nmc Pool p_lif Field spikes gt gt mmc record nmc Volume 6 11 8 3 1 Field Vm gt gt mmc record nmc Volume 10 1 1 12 3 1 Field Vm 2 4 3 Running the simulation Now we are ready to run the simulation lets say for 1 sec gt gt R simulate nmc 1 S This returns the cell array R which contains the response of the network See the section about input and output Sec 3 for more details about the structure of R circuit response l l l l l 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 fl li J 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 jl jl il 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 time sec Figure 7 Response However one can use the command gt gt plot_response R to plot the response R This results in the plot shown in Figure 3 Stimulus and Response 3 1 The Stimuls or Input Signals When runing a network simulation via simulate nmc Tstim S one can specify the stimulus D S has to be a struct array with the following fields e S channel i spiking binary flag 0 1 which determines if S channel i data should be interpreted as spike tim
12. yan P and Abbott L 2001 Theoretical Neuroscience Computational and Mathematical Modeling of Neural Systems MIT Press See also http people brandeis edu abbott book Gerstner and Kistler 2002 Gerstner W and Kistler W 2002 Spiking Neuron Models Cambridge University Press See also http diwww epf1l ch gerstner BUCH html 11
Download Pdf Manuals
Related Search