SpiNNakerManchester / sPyNNaker

The SpiNNaker implementation of the PyNN neural networking language
Apache License 2.0
105 stars 44 forks source link

Different post-synaptic neuron have same weights with lateral inhibitory #207

Closed hh2712 closed 8 years ago

hh2712 commented 8 years ago

Hi, I have constructed a two layer neural network with STDP mechanism and leaky integrate and fire model. For each neuron in the second layer, it imposes inhibitory connection to all other neuron on second layer.

I have then used getWeights() function to retrieve the weights. The post-synaptic neuron do behave as I expected, but there are neurons having exactly the same weights, even I have implemented the lateral inhibitory.

Could anyone kindly give me some suggestions on this? I have attached the relevant code below.

-------------------------------------------------------------------------------------------------------------

connectionsOn = sim.Projection(ON_pop, post_pop, sim.AllToAllConnector( weights = weight_ini,delays=1), synapse_dynamics=sim.SynapseDynamics( slow=stdp_model))

inhibitory between the neurons

connection_I = sim.Projection(post_pop, post_pop, sim.AllToAllConnector(weights = 5,delays=1,allow_self_connections=False), target='inhibitory')

sim.run(simtime) weights_trained = connectionsOn.getWeights(format='array') sim.end()

-------------------------------------------------------------------------------------------------------------

Many thanks, Hanyi

AlexRast commented 8 years ago

We need more information please to help you on this.

From the code snippet and what you are describing it's difficult to tell what, if anything, is going wrong.

Can you describe what you would expect to see and why? Describe your network and how you're setting it up so we have the context. Can you also please give us a high-level description of what the network is supposed to do?

Also please send the full PyNN script and then it should be easier to identify what's happening.

Thanks.

On 13/04/16 12:06, hh2712 wrote:

Hi, I have constructed a two layer neural network with STDP mechanism and leaky integrate and fire model. For each neuron in the second layer, it imposes inhibitory connection to all other neuron on second layer.

I have then used getWeights() function to retrieve the weights. The post-synaptic neuron do behave as I expected, but there are neurons having exactly the same weights, even I have implemented the lateral inhibitory.

Could anyone kindly give me some suggestions on this? I have attached the relevant code below.

-------------------------------------------------------------------------------------------------------------

connectionsOn = sim.Projection(ON_pop, post_pop, sim.AllToAllConnector( weights = weight_ini,delays=1), synapse_dynamics=sim.SynapseDynamics( slow=stdp_model))

inhibitory between the neurons

connection_I = sim.Projection(post_pop, post_pop, sim.AllToAllConnector(weights = 5,delays=1,allow_self_connections=False), target='inhibitory')

sim.run(simtime) weights_trained = connectionsOn.getWeights(format='array') sim.end()

-------------------------------------------------------------------------------------------------------------

Many thanks, Hanyi

— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/SpiNNakerManchester/sPyNNaker/issues/207

alan-stokes commented 8 years ago

Can I ask that you move this discussion into the google group (https://groups.google.com/forum/#!forum/spinnakerusers). This is because I'd prefer the entire community (rather than just the software members [who follow spynnaker]) to help diagnose if this is either a misunderstanding on PyNN / how models work vs an actual software bug.

Therefore I'm closing the issue with the caveat that if it indeed shows some flaw in the software, it can be reopened.

hh2712 commented 8 years ago

No problem.

Thanks, Hanyi

Sent from my iPhone

On 14 Apr 2016, at 12:19, Alan Stokes notifications@github.com wrote:

Can I ask that you move this discussion into the google group (https://groups.google.com/forum/#!forum/spinnakerusers). This is because I'd prefer the entire community (rather than just the software members [who follow spynnaker]) to help diagnose if this is either a misunderstanding on PyNN / how models work vs an actual software bug.

Therefore I'm closing the issue with the caveat that if it indeed shows some flaw in the software, it can be reopened.

— You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub

hh2712 commented 8 years ago

Hi, I have re-posted my question on google group again, with a clarified background information and attached code underneath.

Would it be possible for you to help me look into this issue again?

Thanks and regards, Hanyi

On Thu, Apr 14, 2016 at 1:40 PM, Google Mail hanyi.hu12@gmail.com wrote:

No problem.

Thanks, Hanyi

Sent from my iPhone

On 14 Apr 2016, at 12:19, Alan Stokes notifications@github.com wrote:

Can I ask that you move this discussion into the google group ( https://groups.google.com/forum/#!forum/spinnakerusers). This is because I'd prefer the entire community (rather than just the software members [who follow spynnaker]) to help diagnose if this is either a misunderstanding on PyNN / how models work vs an actual software bug.

Therefore I'm closing the issue with the caveat that if it indeed shows some flaw in the software, it can be reopened.

— You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub https://github.com/SpiNNakerManchester/sPyNNaker/issues/207#issuecomment-209887388