Hi Chandan,

Thanks for sharing this interesting article. There are a number of papers
reporting high probability of bidirectional connections in the cortex. But
this "high-probability" is relative to a null hypothesis that assume
independent random connections among pairs of neurons. If you look at the
absolute numbers, it is still safe to say that the majority of the
connections are unidirectional. I am not surprised that connections in the
cortex is more organized than a random network, but we cannot conclude from
this paper that bidirectional connection is the general principle in the
cortex.

--
Yuwei Cui

Research Engineer, Numenta Inc.

LinkedIn: https://www.linkedin.com/pub/yuwei-cui/1b/400/866

On Mon, Aug 3, 2015 at 10:49 AM, Chandan Maruthi <[email protected]>
wrote:

> Pascal,
> You are right , synaptic connections are unidirectional, however there
> seems to be a high probability of bidirectional connections. at different
> distances
>
> Found an interesting article on synaptic connections .
> http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1054880/
>
>
>
>
> On Sun, Aug 2, 2015 at 2:44 AM, Pascal Weinberger <
> [email protected]> wrote:
>
>> Well that's how it is in biology as well...
>> Despite of gap-junctions all synapses are one directional.
>> This is also logically sensical, as one feature can predict another while
>> the other does not necessarily mean the first one is present. You want to
>> learn these connections separately.
>>
>> On a high level you may draw the analogy that 'Car' implies 'seat'; but
>> 'seat' does not necessarily  (or with the same strength) imply 'car'...
>> On Aug 2, 2015 4:26 AM, "Chandan Maruthi" <[email protected]>
>> wrote:
>>
>>> Is a Nupic synaptic connection 2 way? Or is it one way by default and
>>> has to explicitly be trained to learn a new connection back.
>>>
>>>
>>> So cell a on column a1 connected to cell b on column b1 does not
>>> automatically mean activation of b will result in prediction of a right?
>>>
>>> I see why it is so, but it seems inefficient , doesn't it?
>>>
>>>
>>>
>>> --
>>> Regards
>>> Chandan Maruthi
>>>
>>>
>>>
>
>
> --
> Regards
> Chandan Maruthi
>
>

Reply via email to