Wednesday, March 21, 2007

Numenta and Hierarchical Temporal Memory

I read an article in Wired and due to varous interests i have it really caught my attention. It would around the next step in intelligent software called Hierarchical Temporal Memory.

Quite, erm simply...

Hierarchical Temporal Memory (HTM) is a new computing paradigm that replicates
the structure and function of the human neocortex. HTMs can perform tasks which
have been easy for people but hard for computers - for example, recognizing
objects, making predictions, understanding speech, or navigating in a complex
environment. HTMs promise to approach or exceed human performance in many
cognitive tasks.
I ran their demonstration software and it was pretty good - it really only stopped working at the point where i myself wouldn't really have recognized the image. The first test i did worked fine, so i added some noise to test what would happen. The successful result is shown below:

I kept adding noise to see when i would actually break it and the noise tipping point is shown below.

The whole concept is excellent. I read Emergence and various other titles and this kind of thing was in my head. I knew this is the only way it could work, but where to start was the problem. Well, Numenta have started and it's pretty cool stuff. What i really like about it is that it involves community and collaboration - not in the Web 2.0 style, but rather each layer talks to each other and feeds back - the way we think and learn. It's pretty exciting. The work i am doing is more along the lines of peer collaboration, but i can see so many similarities on how i think these things can work that i just kinda know this is the way this stuff should be. Start simple and layer complexity.

No comments: