Would it be possible to simulate deep spiking neural network with convolutional structure using Brian 2

[Xu Zhang]

Would it be possible to simulate deep spiking neural network with convolutional structure using Brian 2? Any examples?

[Dan Goodman]

It's certainly possible and I'm pretty sure people are doing it, but you
might find it hard to find example code because it's pretty new stuff.

Dan

On 03/03/2017 00:06, Xu Zhang wrote:
> Would it be possible to simulate deep spiking neural network with
> convolutional structure using Brian 2? Any examples?
>

> --
> http://www.facebook.com/briansimulator
> https://twitter.com/briansimulator
>
> New paper about Brian 2: Stimberg M, Goodman DFM, Benichoux V, Brette R
> (2014).Equation-oriented specification of neural models for simulations.
> Frontiers Neuroinf, doi: 10.3389/fninf.2014.00006.
> ---
> You received this message because you are subscribed to the Google
> Groups "Brian" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to briansupport...@googlegroups.com
> <mailto:briansupport...@googlegroups.com>.
> To post to this group, send email to brians...@googlegroups.com
> <mailto:brians...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/briansupport.
> For more options, visit https://groups.google.com/d/optout.

[Marcel Stimberg]

To add to Dan's reply and to be clear: if you are interested in neural
networks purely for machine learning, then Brian is not the tool for you
-- it is a simulator for *biological* neural networks. If you are
interested in using biological neural models for deep learning and want
to express not-commonly used models, then it *might* be a useful tool.
Either way, do not expect it to scale to networks of millions of
neurons, to be GPU-accelerated and all that -- Brian is not at all a
tool comparable to tensorflow.

Best,

Marcel

[Dan Goodman]

Agree absolutely that Brian is not a spiking neuron equivalent to
tensorflow, but I think since he was asking specifically about *spiking*
deep neural networks then it would be an appropriate tool at least in
the early stages. It doesn't necessarily have to be biological neural
networks, only spiking ones.

Dan

[Marcel Stimberg]

I am still not convinced :) Implementing stuff like back-propagation,
convolutional layers, etc. *is* possible in Brian, but you have to hack
stuff together using some Brian internals (e.g. we discussed an
implementation of shared weights previously:
https://groups.google.com/d/msg/brian-development/7n7SXLUt0mY/5o5iCNJFPR4J)
and in general you have to work around the limits of Brian while not
profiting much from its strengths. If you are already very experienced
in Brian and/or have a clear idea for an implementation in "biological
terms" (membrane potentials, delays, synapses, thresholds, resets,
etc.), then go for it.

Either way, if anyone on the list has done anything in that direction
with Brian, I'd be certainly interested in hearing about it!

Best,

Marcel

[Xu Zhang]

I see. Thanks for Marcel's and Dan's reply. In the deep spiking neural network field, I found some codes using Torch and others using Matlab. They all program for SNN themselves. I don't know whether it would be possible to using Theano to train a ANN and mapping it to a Brian simulated SNN. I'll check whether it would be possible to build convolution layers in Brian using spiking neurons. I guess the structure has to be 3D and use subgroup.

"Fast-Classifying, High-Accuracy Spiking Deep Networks Through Weight and Threshold Balancing"