Dear Rolandas,
Maybe this is really more a "pyNN" question than a "hardware" question.
Typically, a SpikeSourceArray is not "a neuron that spikes"; rather it
is an abstract object that generates spike events at predefined spike times.
In the case of the Spikey chip, these spikes are fed into the system at
the predefined spike times via a subset of synapse drivers. The first
SpikeSourceArray generated in the pynn script will send its spikes
through driver no. 255 (resp 511), the second SpikeSourceArray will use
driver 254 (resp 510),... For the numbering please see
https://electronicvisions.github.io/hbp-sp9-guidebook/pm/spikey/spikey_school.html#introduction-to-the-hardware-system
With this in mind, I must admit that I don't fully understand your question:
"I want to create a network of let's say 100 neurons connected all to
all, that gets input from other 10 neurons. Now, from time to time I
want to use different 10 neurons (to train different patterns). The
question is, how to address which 10 neurons on the chip to take?"
Maybe what you have in mind are more than 10 "other neurons"? Are these
maybe multiple populations, each of size 10? Are these inputs
full-fledged neurons that have a membrane potential, reversal
potentials, etc.? Or is their only distinct property their spike time?
In this case, maybe you could generate multiple populations of
SpikeSourceArray's, and let only one of the populations fire per
training epoch?
Are these ideas pointing in the right direction? Otherwise I can offer
that you send me a sketch of your network architecture, or a minimal
pyNN script that works properly in NEST. Maybe this can help to shed
light on this matter.
Best regards,
Johannes
On 11/09/2016 11:48 PM, SUBSCRIBE KIP-SPIKEY-USERS Anonymous wrote:
> Dear Johannes,
>
> thanks for your response.
>
> It is probably that I don't understand something very simple. My problem is that I want to create a network of let's say 100 neurons connected all to all, that gets input from other 10 neurons. Now, from time to time I want to use different 10 neurons (to train different patterns). The question is, how to address which 10 neurons on the chip to take?
>
> In something like this:
>
> pInput = pynn.Population(10, pynn.SpikeSourceArray, {'spike_times': np.arange(ststart, simtime, ststep)}, label="input")
> pCausal = pynn.Population(100, pynn.IF_facets_hardware1, neuronParams) #Integrate-and-fire model for use with the FACETS hardware
> ...
> connection_method = pynn.AllToAllConnector(weights=7*wemin)
> prjCausal = pynn.Projection(pInput, pCausal, method=connection_method, target='excitatory', synapse_dynamics=pynn.SynapseDynamics(slow=stdp_model))
>
> I'm not sure which 10 neurons on the chip it is using and how to pick other 10?
>
> Hope that my questions make sense.
>
> Thanks,
> Rolandas
>
> ########################################################################
>
> To unsubscribe from the KIP-SPIKEY-USERS list, click the following link:
> http://listserv.uni-heidelberg.de/cgi-bin/wa?TICKET=NzM2MzA3IGJpbGxAS0lQLlVOSS1IRUlERUxCRVJHLkRFIEtJUC1TUElLRVktVVNFUlMgIAqwKna7hI0y&c=SIGNOFF
>
########################################################################
To unsubscribe from the KIP-SPIKEY-USERS list, click the following link:
https://listserv.uni-heidelberg.de/cgi-bin/wa?SUBED1=KIP-SPIKEY-USERS
|
