Different Spatial Pooler algorithms

Hi all,

I have found in some paper(s) (I.e.:“Application of Numenta® Hierarchical Temporal Memory for land-use classification” http://www.scielo.org.za/scielo.php?script=sci_arttext&pid=S0038-23532009000500014 )
that there are different SP algorithms: Gaussian, Dot and Product.
I’m familiar with the JAVA implementation of SP and didn’t find anything related to different SP algorithms.
Can anybody explain this?


I think that’s the old version of HTM.
Today’s HTM is a completely different algorithm.


Agree, we haven’t use the term Temporal Pooler for ages (besides when we actual meant it). Looks like something when Numenta transforming from the Zeta algorithms to CLA.

For reference:

Old terminology New terminology
Temporal Pooler Temporal Memory
Spatial Pooler Spatial Pooler
N/A Temporal Pooler - A TM that can learn sub-sequences
CLA (Cortical Learning Algorithm) HTM (Hierarchical Temporal Memory)
1 Like

@rhyolight Just for curiosity, do you have whitepapers/documents from the era of the paper? I’m curious about HTM’s history.

We only have git history from NuPIC.

@marty1885 Subutai presented a review of older Numenta tech (Zeta1, CLA) in this video: https://www.youtube.com/watch?v=6_wattbWgiU