[nengo-user] Interpreting Value Graph

[Terry Stewart]

Hi Brian,

The value graph is a function of the spike pattern coming out of the
group of neurons, not the voltages.  The idea is that it's a
representation of "if you had inputs from all the neurons in that
group, what would be your best estimate of the actual value being fed
into it".  Since the outputs from neurons are spikes, we use the
spikes rather than the voltages.

As for how we compute it, we take the spikes from each neuron and do a
linear weighted sum to get the value.  The weights we use for each
neuron is what we call a "decoder", and we simply find the optimal
decoder (i.e. the one that gives the best value estimate).  There's an
overview of this algorithm here
<http://compneuro.uwaterloo.ca/files/publications/stewart.2012d.pdf>
and the equation for computing the value estimate is equation 2.

I should also note that we do not use firing rates.  We use actual
individual spike times for all of this.  The only change when using
spike times is that we have to note that a spike is not an
infinitesimally quick event.  Rather, the effect of a spike is that it
generates an input into the next neuron that lasts over time (and the
amount of time varies from ~2ms up to ~200ms depending on what
neurotransmitters are involved).  So all we do is replace each spike
with its generated "post-synaptic current" and continue to use this
decoder approach.

Does that help?

Terry

On Wed, Oct 1, 2014 at 6:20 PM, Brian Krainer <bkrainer731 at gmail.com> wrote:
> How exactly is the value graph calculated? Is it supposed to be
> representative of the neural firing rates, or the voltages?
>
> Thanks,
>
> Brian
>
> _______________________________________________
> nengo-user mailing list
> nengo-user at ctnsrv.uwaterloo.ca
> http://ctnsrv.uwaterloo.ca/mailman/listinfo/nengo-user
>