I don't understand, doesn't the current NuPIC include a temporal pooler?
What about all the TP configurations within a CLA model?

Pedro.


On Tue, Jan 7, 2014 at 5:17 PM, Jeff Hawkins <[email protected]> wrote:

> Ben,
> "Temporal Pooling" is an important concept in brain function and cortical
> models.  It says that the cortex takes a series of patterns that are
> adjacent in time and forms a stable representation for the series.  The
> cortex uses adjacency in time to know that different patterns actually
> represent the same thing and therefore should have a common representation.
> The concept of temporal pooling is discussed in the HTM/CLA whitepaper.  We
> also have an implementation of temporal pooling discussed in the white
> paper
> that we tested a few years back.  There are several problems with the
> temporal pooling mechanism in the white paper which I have discussed with
> individuals but are not documented.
>
> The current implementation of the CLA in NuPIC does not include temporal
> pooling (at least it is not enabled).  The CLA today can learn sequences
> but
> it doesn't have the ability to form stable representations of sequences,
> that requires temporal pooling.
>
> Over the last month I have made progress in better understanding temporal
> pooling, I believe I know how it works in the cortex, and how to address
> the
> issues with the old implementation.  This is something I hope to write
> about
> in the next month or two and get someone to test over the coming year.  I
> think it is too complex a topic to discuss fully during an office hour.  My
> plan
>
> Jeff
>
> -----Original Message-----
> From: nupic [mailto:[email protected]] On Behalf Of Benjamin
> Wiley
> Sent: Tuesday, January 07, 2014 4:56 AM
> To: [email protected]
> Subject: Re: [nupic-discuss] (no subject)
>
> sorry for this newbie question, but when people respond to Matt's question
> "What is the most important work that needs to be done for NuPIC this
> year?"
> with "Temporal Pooling", what do they mean exactly?  What is it about the
> temporal pooling that needs to be improved?
>
> Thank you,
>
> Ben
> _______________________________________________
> nupic mailing list
> [email protected]
> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>
>
> _______________________________________________
> nupic mailing list
> [email protected]
> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>



-- 
Pedro Tabacof,
Unicamp - Eng. de Computação 08.
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to