Nice poster,

thanks for sharing, Matt!

I have a couple of questions if the author(s) are on the list, first the
technical ones:
Could you share the full paper, please?
Is it based on NuPIC, or other HTM/CLA implementation? Will you share the
code?

And a question to the concept, sorry if I completely misunderstood, but the
poster claims it's a "cognitive extention to the HTM/CLA". Can you
elaborate more on the differences to NuPIC's HTM/CLA? What I understand
from the poster is: you wire the HTM in a specific way - such that there's
a top-down input for each layer; and then the error is added to the
bottom-up information at each layer and propagated as/with the SDR. Now, I
think this all falls in the capabilities of the HTM, CLA we have in the
current form (?).

Also if you could elaborate more on the ideas that led you to the
>>top-down with error propagation<< principle? Is there a biological
support? (I think partially, definitely for the TD-flow of information, and
not much for the error propagation?), and inspiration from ML? (looks like
back-propagation from MLP in some way..)

It would be interesting to see the benchmarks scale to more layers.

For the long term goals of LIDA are much aligned with the goals of NuPIC,
so I hope we can collaborate in the future!

Nice work, thanks,
Mark




On Tue, Dec 17, 2013 at 7:17 PM, Matthew Taylor <[email protected]> wrote:

> I think Ryan McCall is on the list, and I saw this on twitter, so
> thought I would share.
>
> https://twitter.com/rmccall84/status/411613318688677889
>
>
> http://ryanjmccall.com/wp/wp-content/uploads/2013/12/mccall-acs-poster-2013.pdf
>
> ---------
> Matt Taylor
> OS Community Flag-Bearer
> Numenta
>
> _______________________________________________
> 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

Reply via email to