Thursday, March 25, 2010

Hawkins explains next generation HTM algorithms

On March 18, Jeff Hawkins gave a talk to the computer science department at the University of British Columbia. In that talk, for the first time, he gave a detailed explanation regarding the upcoming HTM algorithms. To be honest, parts of it were difficult for a layperson (like me) to fully understand. It is a very interesting talk, though, and showed that Numenta is getting ever closer to a truly intelligent computer. This version of the algorithms appears to much more closely mimic the brain's method of learning than ever before. Here is a link to the speech:

http://www.youtube.com/watch?v=TDzr0_fbnVk

6 comments:

  1. Great link! Looks like they are really on to something!

    ReplyDelete
  2. Yes, I get the feeling that they have now cleared a major hurdle toward brain-like computers. It will be interesting to see when they start to demonstrate some real world examples of how well the new technology works.

    ReplyDelete
  3. Too bad the video is made like that, only a small window where you can see Hawkins and the rest for text. Also the sound is of low quality.
    Nevertheless, I will watch it a bit later, at least until I'm totally lost :-)

    ReplyDelete
  4. It's great to see that Jeff Hawkins is doing well. I haven't heard much about him for a few years. I read his book "On Intelligence" 4 years ago. It helps a lot to have read that, before watching the video.

    The last 2/3's of the video is really difficult to understand, but for me the main point is Hawkins is not stuck with his theory, but it is progressing fine. There seems to be a lot more work to be done though, before the model can do anything close to what a human brain can do.

    ReplyDelete
  5. Martin:

    It probably depends on how you look at it. I think that Numenta is actually getting close here with regard to the basic way that the neocortex learns about the world. Hawkins is apparently so excited about this new software model because it appears to do what the brain does (store memories in a sparse distributed manner) while adhering to all of the many known anatomical details of the neocortex.

    There are still many things done by the real cortex not explained in that talk (attention, prediction, specific timing, etc) that Numenta is known to be working on, but if Numenta really has figured out the most difficult part of the whole thing (discovering causes in the world through pattern and sequence learning), it is a big step in the right direction.

    ReplyDelete