This was a very interesting read that hopefully some of the Numenta skeptics will take a look at closely. A few points that I pulled out of the paper were as follows:
1. The new learning algorithms are very closely tied to the biology of the brain. The new HTM software models the hierarchical levels, columns of neurons, the neurons themselves, and even the dentrites and synapses. Numenta clearly believes that HTM now learns in a similar way as the neocortex.
2. The algorithms appear to be scalable to any size. It sounds like the user can set the number of columns, the number of neurons per column, the number of levels, etc. and the only real limiting factor on scalability is the power of your computer and the amount of memory that you have available.
3. For the first time, prediction is now at the center of the HTM algorithms. On Intelligence, of course, postulated that prediction is at the heart of what the brain does, and is what makes us intelligent, but HTM until now really didn't implement prediction. Now that a more brain-like method is being used for the learning of sequences of patterns, HTM appears to have a powerful prediction mechanism. According to the paper, anytime HTM recognizes that an input is part of a sequence, it will automatically predict future inputs. It can do this based on sequences that go far into the past, and can predict not just one, but a number of time steps into the future. These capabilities will be important when someone decides to use HTM to control a robot, since according to Numenta, prediction and directing motor behavior are very similar activities. For instance, when a robot has a goal to accomplish some task, it will use prediction based on its remembrance of learned sequences that constitute prior motor actions to direct its future actions.
4. A number of theoretical HTM capabilities are not yet implemented in the software. Numenta specifically mentioned attention mechanisms, motor behavior for a robot or some other physical embodiment of an HTM, and specific timing for the learning of sequences that happen at particular speeds (such as music). Still, it will be very interesting to see the acceleration of commercial applications with the significant advance that these algorithms represent.
5. This paper is only a working draft. It was mentioned that several future chapters are planned for the book, including a chapter on the mapping to biology, and a chapter on how the algorithms have been and can be applied to applications.
Here is the link:
http://www.numenta.com/htm-overview/education/HTM_CorticalLearningAlgorithms.pdf
Subscribe to:
Post Comments (Atom)
nice new learning algorithms...i am surely gonna use them
ReplyDeleteJunaid
www.trendyonlinesolution.in
Very interesting paper. What's with Michael Anissimov's belligerence? I think you did a good job presenting your argument to him.
ReplyDeleteNo idea. My first post was a bit of a retort because I felt like his dismissal of Numenta was very cursory, and that didn't change throughout any of the comments on that blog. By the time I read that whole thread, I could only conclude that Anissimov has simply decided to reject Numenta out-of-hand just so he can say that he was right if it turns out that Numenta fails in its objectives.
ReplyDeleteHe is a much brighter guy than me, and he claims to have spent over 100 hours thinking about Numenta's ideas, yet it is patently obvious that he doesn't have a grasp of the depth of HTM's mapping to cortical anatomy. It is ridiculous to claim that Numenta is merely copying a few superficial cortical details. Rather, Numenta has come up with a top-down theory of several things that the cortex must be doing (time-based learning, inference, and prediction) and has spent the last several years trying to figure out how to match those functions with the known cortical anatomy. In other words, this isn't a superficial or "shoot by the hip" approach, but a methodical combining of theory and knowledge into the best possible algorithms.
I'm not sure he's that much brighter than you (if at all) - he just tries to come across as if he is. That's the thing that bothers me about his blog - it is very arrogant sounding, as if he is an expert at all of these things. Then he gets all mad and defensive when people say "what do you know?". If he just sounded a bit less cocky, people wouldn't be so quick to judge his credentials.
ReplyDeleteYes, cocky and without the knowledge to back it up (at least when it comes to Numenta). It is still amazing to me that he claims to have spent 100 hours thinking about Numenta, but hasn't read their seminal peer-reviewed paper on their work.
ReplyDeleteI get the sense that he is one of the philosopher AI types who have difficulty figuring out how sensory perception algorithms like Numenta's can solve the deep mysteries of the brain pondered by philosophers for centuries. Numenta is just the opposite- they see the brain as a physical thing that can be built if we can reverse engineer it. I am not saying that there is no place for the philosophical side of AI, but in terms of making measurable progress, I would take an engineer over a philosopher any day.