Cubic spiking neural network

This is a spiking neural network composed of about 65000 neurons using the CUDA C/C++ extension. 20% of the total amount are inhibitory neurons and the remainings are excitatory. Each neuron has 64 presynaptic neurons and they are places on a 3d cube (here you see the top of it). In this video you can see the pattern formation.

Category:

Science & Technology

Tags:

• neuron
• neurons
• Nvidia
• GPU
• synapse
• spiking neural network
• pattern
• network
• CUDA

[MultiNeurons]

a nanoarray? What is it exactly?

[mikeccuk2006]

what GPU model do you use? can you simulate any more neurons than 65000?

[MultiNeurons]

@mikeccuk2006

I am using a GTX275 with 896M of ram. Let's say it's a second family of the nvidia GPU. Of course it's possible to simulate as many neurons as you want. I did choose 65536 neurons because of technical reasons.

[chromosome24]

what kind of input are you giving it?

[r0galik]

Looks like a nanoarray

[MultiNeurons]

@ElessarSemaj

it's not very difficult indeed. A limitation I adopted is that the presynaptic neurons for every neuron are 64. That is a 65536 x 64 table where every element is the weight of the connection. This can be excitatiory (variable from 0 to 3.0 in this case) or inhibitory (this is alwasy at the same level: -5.0). Hope this can help you...

[ElessarSemaj]

@mrule7404 I'll also give it a go, gonna be pretty difficult though^^

[mrule7404]

I want to try to replicate this.