Hi Roy,

The presynaptic neuron only has one mode of signalling, ie the axon is
either on or off. The inhibition occurs strictly at the synapse; if the
presynaptic axon has a synaptic bouton which produces inhibitory
neurotransmitters, then the postsynaptic activity will be reduced or
eliminated. "Normally" the presynaptic axon has a bouton containing
excitatory neurotransmitters, which raises the postsynaptic potential
towards the threshold.

In the CLA, we normally add contributions from synapses; this reflects the
positive contribution of incoming axons. The sum of "connected" synapses
(those with permanences above the threshold) is calculated, and if it
exceeds a threshold on a dendrite, the cell will fire (or in our case enter
a predictive state). We don't model inhibitory synapses at all.

If we were to use the idea of inhibitory synapses, we would do this by
changing the sign of the contribution for this type of synapse when adding
up the synapses on the dendrite. The permanence only reflects how
well-formed the synapse is, not whether it's excitatory or inhibitory.

This is similar to a proposed algorithm which has been put forward in
another thread, but in that case we were only talking about using
inhibition in the SP. That proposal had high-activity columns causing
neighbouring columns to reduce their activity potentials, causing them to
be less likely to become active. It looks like this new algorithm works
quite well and is more efficient than the current "top 2%" local inhibition
strategy.

[1] https://en.wikipedia.org/wiki/Synapses
[2] https://en.wikipedia.org/wiki/Inhibitory_postsynaptic_potential

Regards,

Fergal Byrne








On Wed, Sep 18, 2013 at 3:46 AM, Marek Otahal <[email protected]> wrote:

> Not really Roy,
>
> inhibition is not "sent" like a signal along axon. It's more like a
> chemical reaction, that happens (strictly) locally (synaptic receptors are
> temporaraly blocked by a chepical, eg GABA). When a certain (membrane)
> potential reaches a level, the neuron reduces/stops its activity -- imagine
> a leaky bucket that refuses to work when full, until the water drops..
> This blocking mechanism is in place to avoid strong "waves" of signal
> occuring in the network.
>
>
> On Wed, Sep 18, 2013 at 4:27 AM, Roy Gu <[email protected]> wrote:
>
>> Hi Fergal,
>>
>> Thanks. I would like to ask whether permanence be negative biologically?
>>     We know the cell can be fired at a 'fire rate' and send the signal to
>> its connected cells. Can the cell sent another type of signal to its
>> connected cells and the signal can be treated as a 'inhibition' and
>> mathematically a negative weighted permanence?
>>
>> Regards,
>> Roy
>>
>> ----
>> Hi Roy,
>>
>> In the CLA, "inhibition" is done at the column level, either globally or
>> locally. The Spatial Pooler picks the 2% of columns with the highest
>> activation potential from either the whole region (global) or sub-regions
>> (local) and activates them. This mimics inhibition in the neocortex, in
>> which high-potential (or most active) cells suppress the potential or
>> activity of cells in neighbouring columns.
>>
>> The overall effect is very similar, in that sparseness is imposed and the
>> "best" columns are activated. If we can achieve very similar outcomes,
>> then
>> efficiency considerations justify the deviation from the actual process in
>> the neocortex.
>>
>> A recent thread on this list discussed an alternate inhibition strategy
>> for
>> the CLA. This proposal involves the idea of high-potential columns
>> imposing
>> a "negative potential" effect on its neighbours. Initial tests suggest
>> that
>> this method is significantly faster than the current local "inhibition"
>> algorithm, and it is also a closer analogue to the neocortex.
>>
>> Regards,
>>
>> Fergal Byrne
>>
>> _______________________________________________
>> nupic mailing list
>> [email protected]
>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>
>>
>
>
> --
> Marek Otahal :o)
>
> _______________________________________________
> nupic mailing list
> [email protected]
> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>
>


-- 

Fergal Byrne

ExamSupport/StudyHub [email protected] http://www.examsupport.ie
Dublin in Bits [email protected] http://www.inbits.com +353 83
4214179
Formerly of Adnet [email protected] http://www.adnet.ie
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to