Code required for a research paper

[Saurav Pawar]

Hello,

I read this research paper and wanted to ask, whether the code for this research paper is avalilable or not as my aim is also to simulate a 5 layered spiking CNN on SpiNNaker ?

The reason I am asking here is because, in the research paper the SNN was simulated on SpiNNaker.

Regards,

Saurav

[Andrew Rowley]

Hi,

 

We don’t have the original code, however we have since implemented something similar to this in the main software.  This is very recently available in the master branch of the software; it should work but there might be some issues as it hasn’t been used that much so far.  Some instructions on the use of convolutions in this way can be seen here:

https://spinnakermanchester.github.io/development/2d_convolutions

 

If you do go ahead with this, please let us know if you find any issues,

 

Andrew :)

 

--
You received this message because you are subscribed to the Google Groups "SpiNNaker Users Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spinnakeruser...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/spinnakerusers/e0243428-bddd-41fd-9bcf-38a7f50ef83en%40googlegroups.com.

[Alberto Patiño Saucedo]

Hi Saurav,

I'm one author of that paper. We simulated a pre-trained LeNet (from Rueckauer's ANN-SNN conversion Toolbox), among other architectures, on a SpiNNaker board donated by the APT Group to us (at the time I was a PhD student at UGTO, Mexico). I still have the code, it's quite outdated, but write to me at saucedo at imse-cnm.csic.es and I'll send it to you.

Bear in mind, that LeNet implementation was very far from optimal as it uses an old version of sPyNNaker to load connectivity matrices that don't exploit weight-sharing. The CNN connections are loaded as FC, implying significant memory usage for the connectivity. There is now a ConvolutionConnector, as Andrew points out, that uses hardware in a more efficient way and should fit better for what you're trying to achieve. 

Best,

Alberto.

[Saurav Pawar]

Hi,

Sure Andrew, thank you. Alberto, I have mailed you on the above mentioned email id, so please check it.

Best,

Saurav