Search for the right paper about HTM

hello Matt,I have just seen a topic “understanding Old Temporal Pooler(2011)”, in there you said the 2011 Numenta white paper about CLA is outdated , and you gave a link where there are several papers on it, but unlike the 2011 version, there seems to be no pseudo code of the whole algorithm which I can learn from , so did you mean that the entire 2011 version paper is unnecessary to read , or does it still have merits in introducin the overal structure of HTM. If not, is there a new version CLA paper which with the pseudocode in it? thank you very much


Hi Student,

That is a good question and I can understand why it might be confusing. It depends a little bit what you are looking for.

  1. This link contains a list of research papers we have published recently on HTMs: These papers are more academic in nature. The first one (Why Neurons Have Thousands of Synapses) is the best one to read to get an overview of the theory and sequences. It is written from a Neuroscience perspective and contains a mathematical description, but no pseudocode.

  2. This link is a new one and contains an evolving textbook on HTMs: It is a “living book” and currently incomplete and in process, but it is the best replacement for the old whitepaper. The first couple of chapters give an overview. For this book, I actually updated the pseudocode of the SP and temporal memory algorithms to make them current, but it is currently buried in there. To get to the pseudocode, go to “Content to be incorporated” and then scroll down to “HTM Learning Algorithms”. I would strongly recommend using this version of the pseudocode as it uses newer terminology and removes some obsolete bits. Don’t use old the whitepaper version.

  3. There are also some good videos of Jeff describing the overall HTM theory here:

I hope this helps!



There’s also this Wiki page on this site (just up today, expect it to grow).


thanks mr Subutai and mr Byrne

1 Like

by the way mister subutai, is there any materials about applying HTM in vision recoginition problems? thanks

Here is all the vision stuff we pulled out of NuPIC awhile back:

hi, thank you Matt, is thie based on Numenta’s latest version algorithm , (that is , there is only temporal memory and not heirarchy in it)?

Actually, it depends on having NuPIC installed, so it uses whatever algorithms is in the nupic installation. I believe it was architected as an extension to NuPIC. Please correct me if I’m wrong, @scott.