Curious amount of HTM papers flying under the raidar

I was searching around for old Zeta1 algorithm related stuff. And I came across this page linking to the original Zeta1 publication. Some clicking around I found a paper by Alex Pappachen James using Zeta1 for facial recolonization. Scroll down the pages you’ll find the authors have published quite a few HTM papers in the past few years. Like this one Using HTM for word recognizition, Analog HTM circut , Neuromemristive Circuits for Edge Computing, etc…

I am quite curious that these papers aren’t brought to the forums more. There’s a healthy amount of practical information about applying HTM to real world problems. Which is what we (as a community) quite struggle at.

But the papers are not of top quality. - I have read the word recognition paper (thanks university VPN). It seems legit-ish. It looks very like what a hardware engineer writes and is quite poor on evaluating the ML part. I guess I trust IEEE enough to give it a pass. The other papers follows the same trend; good hardware talk but you need to squint to see what they are doing in the ML section.

What do you think?

10 Likes

Anything older than 2011 should not be taken too seriously, because it is based on the old “HTM” which was Bayesian.

But those 3 new papers I have never seen! Thanks for posting them! I’ll try to read them…

2 Likes

You can dig deeper into the authors. There’s way more than what I posted. And all of them looks interesting!

2 Likes

Here’s another overlooked HTM paper. Implementing HTM on Adapteva’s manycore Epiphany processors.

1 Like