Hi Marek, While the TP->SP change I recommended in November is a relatively simple but powerful one ([1] is a description plus a cheap hack for NuPIC which accomplishes it), it turns out that this alone is not enough to really leverage the integration between spatial and temporal processing for use cases beyond streaming numerical data (ie those which Grok is optimised for). It's necessary but not sufficient (and, as Grok proves, it's not strictly necessary for certain simple problems).
I've been working on a different kind of design for a HTM-CLA system over the last few months [2]. The plan is to have something which is simultaneously very close to Jeff's vision of the neocortex, and also a real production system for developing solutions. In a nutshell, the ambition is to build a "VisiCalc for Machine Intelligence" (hats to Jeff Fohl for inspiring this catchphrase). [1] http://inbits.com/2013/11/adding-prediction-to-the-nupic-spatial-pooler/ [2] http://fergalbyrne.github.io/ On Mon, Apr 21, 2014 at 1:40 PM, Marek Otahal <[email protected]> wrote: > Hi all, > > I'd like to ask if there are any plans about the new spatial pooler idea > introduced by Fergal, that would use cells and be able to do predictions? > (I couldn't find the old ML thread that I wanted to ping) > > In additions to other advantages, it'd make SP and TP more unified (which > they in practice really are one), and could simplify things for > reconstruction... > > Cheers, Mark > > -- > Marek Otahal :o) > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > -- Fergal Byrne, Brenter IT Author, Real Machine Intelligence with Clortex and NuPIC https://leanpub.com/realsmartmachines <http://www.examsupport.ie>http://inbits.com - Better Living through Thoughtful Technology http://ie.linkedin.com/in/fergbyrne/ https://github.com/fergalbyrne e:[email protected] t:+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
