What would you say... you do here?

Hey Guys,

I was turned on to the ideas of Numenta and HTM about a year ago. A supervisor at my university was saying it was all the rage.

I worked with the architecture for about six months now, followed the community and Numenta as well, and I have a simple question:

What can HTM do at this point?

You all have spent much more time than me in this area, years, I assume, and yet I get this odd feeling that HTM has very little to show for itself…

Perhaps y’all can prove me wrong and post some interesting stuff in threads below this post. I hope you can.

Kind Regards
altshiftj

I’m not the right person to directly answer.

One thing is, the newer stuff is about much more complicated neuroscience. It’s based on neuroscience, often things like “this particular group of cells fires at X locations”. The circuitry to do those things is very hard to understand and study, so it’s hard to figure out exact mechanisms. I think it might be bottlenecked by scientific studies at this point, so we might just need to wait like 10 years. Their papers have been pretty successful though, so hopefully those studies happen. It’s just a slow process.

By my own experience, only stuff related to Sequence Memory: memorizing sequences, predicting the next step in a sequence, anomaly detection (on sequences)

What does it do? Numenta and Numenta’s mission inspire me more than any other brain research I see out there. So I think bringing a community together like this has huge value, and I love it. It is like being part of a research team when you get access to their meeting videos and discussions. I find out about many cool papers on here that I don’t hear in popular science media.

As far as tech, you seem to have a pragmatic engineers mind, asking for the tech specs for this year. But Numenta (correct me if I am wrong here) is focused on 10-20 years out research, so it often bounces around and explores lots of dead ends. This is the nature of open ended research.

That said, there was a post maybe 12-18 months ago where some company used HTM to trade index funds and beat the (at the time bullish) market.