project_presentation

Why Information Flow in A Computer Simulated Cortical
Neural Network Is Important
R. Jarvis Supervisor P. Junor
Department of Electronic Engineering
La Trobe University
July 22, 2015
1 / 31

Introduction
Reverse Engineer Principles of Brain Function: Created a neural
network model of rat cortex including accurate cell shapes. Observed
information ﬂow, making it possible to infer some functions of
individual brain cells.
Made with NEURON-7.3 as a Python module.
100mm3 µm3
http : //neuromorpho.org/neuroMorpho/LSV ideo.jsp
2 / 31

From Reconstructed Cells to Reconstructed Networks.
In order to move from reconstructed Cells to reconstructed networks,
one must connect neurons in a biologically informed way. Spatial and
anatomical data from the Allen Rodent Brain Atlas, is used to inform
cell centre position and synapse polarity.
Neurons are connected by conducting an exhaustive search in the
neural volume for near contact points and allocating a network
connection to the intervening synapse cleft.
3 / 31

From Neurite coordinates to to synaptic cleft
Figure : Two neurons with classiﬁed Membrane with Post Synapses
4 / 31

Wiring Algorithm: Searching the volume of spike sources
HOC
1 proc wire_distances (){
forsec $o1.spk_trig_ls{
3 for (x){
if((x>0) &&(x<1)){
5 if(ismembrane("xtra"){
find_distances (x_xtra(x), y_xtra(x),
z_xtra(x), secname (), x, $o2)
7 }
}
9 }
}
11 }
5 / 31

Self Reﬂection
HOC
1 // String processing performed elsewhere enabled
execution of Strings performed here.
Facilitates network connections.
3 sprint(synapse_post , "%s%s%g%s", sec_string2 , "
syn_ = new AMPA(",storex ,")")
5 execute(synapse_post) //put a post synapse at
the target
7 sprint(synapse_pre , "%s%s%g%s", sec_string1 ,"
nc = new NetCon (&v(",num_seg ,"),syn_)")
9 execute(synapse_pre)
6 / 31

Find Contact Points Algorithm
HOC
forsec $o7.spk_rx_ls{
2 sec_string1=$s4
num_seg=$5
4 if(strcmp(str_src ,str_target)==0){ break }
if(Cell[py.int(tail_src)].div.x[py.int(
tail_target)]>3){ break }
6 if (ismembrane("xtra")) {
for(x){
8 if((x>0) &&(x<1)){
r = (sqrt (( x_xtra(x) - $1)^2 + (
y_xtra(x) - $2)^2 + (z_xtra(x) - $3)^2))
10 if(r <10){// units of um.
if(Cell[py.int(tail_src)].div.x
[py.int(tail_target)]<3){
7 / 31

Pre and Post Synapses
In most what follows I will be referring to pre synaptic locations and
post synaptic locations.
(sources:) http://www.igi.tugraz.at/maass/129/129a.htm
8 / 31

EPSP and IPSP
Some people think of neurons as analog input digital output.
One or more IPSPs can prevent a spike from occurring, by depressing
the membrane potential.
(sources:) http://www.cerebromente.org.br/n12/fundamentos/neurotransmissores/neurotransmitters2.html
http://www.studyblue.com/notes/note/n/chapter-48-nervous-system/deck/4169450
9 / 31

Pre Processing Signals: Spike Trains
Continuous membrane potential thresholded, the times each neuron
ﬁres is stored. Time binning makes signal coarse grained.
10 / 31

Connectivity and the Adjacency Matrix.
I→E=2073, fb1− = 1081
11 / 31

Information In the Spike Train
The number of spikes per bin: ∆t is the source of uncertainty in the
spike train.
There are two major sources of information:
1st the external input (sensory).
2nd sources of variability that are intrinsic to the brains dynamics.
12 / 31

An Interpretation of Information Flow:Prediction
Using Transfer Entropy we can say that neuronY inﬂuences neuronX
if neuronY ’s past activity reduces the uncertainty about neuronX ’s
future activity.
neuronY neuronX
When a particular type of entropy (uncertaintity) is reduced
prediction is increased.
13 / 31

Resulting Information Flow
One cell can predict another cells activity acting through intermediate
neurons
A reduction in ﬁring probability still represents an information
transmission inﬂuence.
Connectivity nTE
14 / 31

Information Flow Between Neurons
Degree matrix shows the quantity and direction of information
transmitted between each pair of neurons.
nTE Correlation Matrix
Presence of information ﬂow contributes validation of reconstruction
because when information does ﬂow, it suggests information is not
randomly created and destroyed at every node.
15 / 31

Information Flow
In Von Neuman architecture no conﬂict between transmitting,
translating, and storing information. Respective tasks performed by
the bus, the CPU, RAM. One entire neuron simultaneously engages in
information transfer, information alteration and information storage.
Sums, Cells[0,24] Sums, Cells[25,49]
16 / 31

Summary
Discussed how TE identified functions. TE distinguished information
sending neurons from information receiving neurons.
Transfer Entropy was used to quantify Information Flow within and
between neurons.
I described how information flow in IPSPs can be detected.
I have discussed how the presence of information flow contributed to
the validation of model.
Generally I have described the analysis of Information flow in a
cortical neural network, and why it is of interest.
17 / 31

References
Kerr, Cliff C et al (2012)
Electrostimulation as a prosthesis for repair of information flow in a
computer model of neocortex
IEEE 20(2), 153 – 160.
Neymotin, Samuel A et al (2011)
Synaptic information transfer in computer models of neocortical
columns
Journal of computational neuroscience 30(1), 69 – 84.
Gourévitch, Boris and Eggermont, Jos J (2007)
Evaluating information transfer between auditory cortical neurons.
Journal of Neurophysiology 97(3), 2533 – 2543
Lizier, Joseph T et al (2012)
Local measures of information storage in complex distributed
computation
Information Sciences 208, 39–54
18 / 31

The End
made with LATEX
19 / 31

project_presentation

More Related Content

What's hot (20)

Similar to project_presentation (20)

project_presentation