Jeff posted an initial version of his new ideas to the list. The title was something along the lines of New Temporal Pooling Proposal.
It does not change the way sequence learning works. It changes temporal pooling to happen between layers rather than as part of the sequence memory. And it incorporated motor input (not output yet) in addition to high order sequence memory for prediction. On Feb 24, 2014 11:43 AM, "Christian Cleber Masdeval Braz" < [email protected]> wrote: > > Hi Ruaridh. > > Thank's for the feedback! Do you know if the new proposal will have a > "sequence learner" and, if so, the implementation will be different from > the actual described in the paper? Or maybe the new TP will be the actual > (sequence) + an improvement. > > > > Regards, > > Christian > > > > 2014-02-22 12:58 GMT-04:00 Ruaridh O'Donnell <[email protected]>: > >> Hello, >> I've just read through your document. It was a nice explanation of the >> details of the CLA algorithm. The diagrams were very helpful at summing up >> the points you made in the text. >> I didn't spot any mistakes, everything looks correct. I'm quite familiar >> with the CLA so I think I might be at least somewhat qualified to proof >> read. One thing you might want to note is that the temporal pooler you >> described (which is the same one as in the white paper) isn't quite how the >> one in NuPIC works. >> The NuPIC one is slightly simpler version that doesn't do as much. It >> only does "sequence memory" as opposed to "temporal pooling". There are >> plans to update the CLA to include temporal >> pooling<https://github.com/numenta/nupic/wiki/New-ideas-About-Temporal-Pooling>but >> that means that the temporal pooling aspects of the white paper >> algorithm will likely become obsolete. Just noting this in case you were >> wanting to continue working on the explanation. It might be better to >> explain the current NuPIC version rather than the white paper version. >> >> Ruaridh >> >> >> On Sat, Feb 22, 2014 at 12:15 PM, <[email protected]> wrote: >> >>> This is great!! Thank you!! >>> >>> Sent from my iPhone >>> >>> > On Feb 20, 2014, at 1:52 PM, Christian Cleber Masdeval Braz < >>> [email protected]> wrote: >>> > >>> > >>> > Hi all. >>> > >>> > I made an attempt of writing a summary of the CLA theory and >>> algorithm. To do that I used the articles, videos, this discussion list and >>> other external materials. As the content is based on my understandings, i >>> would appreciate a lot if the comunity could validade it with corrections >>> and sugestions. >>> > >>> > The PDF follows attached. >>> > >>> > >>> > Regards, >>> > >>> > >>> > Christian >>> > >>> > <IntroductiontoCLAAlgorithm.pdf> >>> > _______________________________________________ >>> > 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 >>> >> >> >> _______________________________________________ >> 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 > >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
