Interesting Scott, SP can produce more no. of SDR bits than its input. Like also, more no. of SDR input to SP could produce less no. of output bits. we could experiment this divergence, convergence characteristics.
Thanks Ramesh Ganesan. On Wed, Aug 7, 2013 at 9:49 PM, Fergal Byrne <[email protected]>wrote: > > Cheers Scott, > > Francisco, the SP is an artefact in the sense that it is a critical > component of NuPIC which replaces the functionality in the neocortex which > causes an SDR of columns to activate in response to feedforward input > (which may not necessarily be SDR in character). (ie it's not an artefact > in the sense of your appendix!). > > This is done in the SP by scoring, choosing the top 2%, updating > dendrites, and boosting (thanks Scott). This enforces an SDR of activated > columns, regardless of the sparsity of the input. It also develops the > learning of input patterns mapped to activation patterns. > > The idea is that the SP "normalises" the input, so that the "top view" > of every region looks similar - a sequence of SDR's. As Scott says, this > "top view" will give no clue as to the character or size of the feedforward > input patterns. > > Regards, > > Fergal Byrne > > _______________________________________________ > 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
