Paper on new learning algorithms now available on Numenta's website

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

New Whitepaper, etc.

The news keeps coming lately from Numenta. Their most recent Twitter post says that the new whitepaper (replacing the original one from March 2007) that describes the new algorithms is going to be published before the end of the month.

Also, I ran across an interesting exchange between a Numenta defender and Ben Goertzel. I am sure you can guess which side of the argument I am on:

http://knol.google.com/k/angelo-c/opencog-numenta-and-artificial-general/1luetnln973wm/3#

Jeff Hawkins speech on November 12

For the first time in eight months, it looks like Jeff Hawkins will be speaking publicly about Numenta's work. The bad news is that, per the contact for the event, it will not be broadcast or recorded. I am hoping that that information turns out to be incorrect. Hawkins is giving the 2010 SmithGroup Lecture at the University of Illinois at Urbana-Champaign. Based on the abstract provided by Numenta, Hawkins will discuss Numenta's overall neocortical theory, the new learning algorithms, and how he believes hierarchical learning machine intelligence models will evolve in the future. Perhaps we will get lucky and someone will create an amateur video of the event. Definitely sounds interesting.

UPDATE: Interestingly, it looks like Hawkins is going to be giving the same talk at MIT's Center for Biological and Computational Learning on November 10. This is great to see, because MIT's CBCL has some of the leading research going on in the field of biologically inspired machine learning. For instance, Tomaso Poggio, who I have mentioned on this blog, is a part of CBCL. This is the type of publicity that I was hoping to see for Numenta. Here is a link:

http://www.facebook.com/CBCL.MIT#!/event.php?eid=150521674991840

DARPA and hierarchical temporal memory

In a recent comment, I linked to an article entitled "Deep Machine Learning- A new Frontier in Artificial Intelligence Research.".

In it, the creators of the DESTIN architecture, who I have mentioned before in this blog, attempt to summarize the field of work going on with Deep Machine Learning, or the idea of using hierarchies to learn in a more brain-like manner. What was interesting to me about the article was that it mentioned a DARPA research effort involving deep machine learning architectures. In April 2009, DARPA put out a request for proposals for deep machine learning architectures. The military is increasingly worried about the vast amount of data that it collects that goes unanalyzed due to the sheer volume of data that humans do not have the time to analyze. DARPA seeks an HTM-like algorithm that will find patterns in this vast data. The DARPA announcement closed in April 2010, and to my shock, I don't see any indication that Numenta put in a proposal (among others, it appears that the DESTIN folks did). In a briefing, DARPA set out a list of desirable properties that would be features of the algorithms resulting from the multi-year research effort. Here is the list:

1. Single learning algorithm and uniform architecture for all applications

2. Unsupervised and supervised learning, including with attention mechanisms

3. Increasingly complex representations as you ascend the hierarchy. Sparse representations were mentioned here

4. The ability to learn sequences and recall them auto-associatively

5. Recognize novelties at each level and escalate them in the hierarchy

6. Feedback for predictions to fill-in missing input

7. Online learning

8. Parameters set themselves and need no tweaking

9. Neuroscience insights to inform the algorithms and architecture

Essentially, that list of desirable features in DARPA's envisioned software is a description of the HTM algorithms. Its difficult to imagine why Numenta didn't throw their hat in the ring given the amount of money potentially involved if the technology catches the eye of the military. In any event, DARPA's document was very interesting reading.

The complexity of the problem faced by Numenta

I have been following Dileep George's new blog, and he made a couple of responses to posts by myself and Dave (perhaps the same Dave who occasionally posts here).

In my post, I asked Dileep how the traditional tree-shaped hierarchy can explain the vast number of qualities that can come into play when we recognise, for instance, a shoe. For example, when we see a shoe, we recognize that it is a certain color, a certain texture, has a certain design on it, and many other features. In other words, recognizing one object requires the brain to have connections to a number of other invariant representations of different types of objects and concepts. In my mind, I couldn't see how a simple tree structured hierarchy could represent this complexity, and Dileep confirmed that I was correct, stating that the brain likely has a number of different hierarchies that communicate with one another. Since that time, I will say that I think that I was mixing up how we recognize a particular instantiation of a shoe with how we recognize the invariant representation of "shoe" that is stored in the brain. The simple tree-shaped hierarchy might be sufficient to store the invariant "shoe" concept while not working to recognize a particular shoe.

Dave's question to Dileep focused on whether a single HTM network could recognize both an object (such as a shoe) and an action (like running or walking). Surprisingly to me, Dileep answered that you would need two separate HTM networks to handle two separate types of knowledge. My conclusion how is that the simple traditional tree shaped hierarchy is not sufficient even to represent all invariant concepts known by the brain, much less the particular instantiations of those representations that we learn (i.e. particular faces of persons as opposed to the general idea of "face").

This goes to show that even if Numenta's new algorithms have licked the problem of how the brain learns within a region and does inference, figuring out how the brain as a whole learns many types of objects and concepts, both invariantly and specifically, and how it ties all of this knowledge together in the amazing way that our brain works, is still something that we are only beginning to figure out.

Numenta's new website

Numenta redesigned its website. Here are a few nuggets from the new site:

1. Some new videos were added, including Hawkins' 2008 keynote from the HTM workshop and a speech by Subutai Ahmad from the 2009 workshop. Ahmad's talk was particularly interesting because he discussed a number of corporate partnerships and some early results from them. For instance, Numenta is/was working with a major automaker on the creation of a pedestrian detection system, where the car looks for pedestrians in front of the vehicle. The early testing resulted in 96-97% accuracy, or closer to 99% accuracy if one counts a false positive as a good result (situations where the system detects a pedestrian where there wasn't one). The talk also mentioned some interesting work that Numenta did with Tyzx, which provides computer vision systems. They used an HTM network to look for objects/persons in security camera footage. Subutai specifically mentioned robotics as a potential application. Interestingly, only three days ago Tyzx announced a deal with Irobot to provide vision systems for its military robots, including person detection capabilities. The press release did not mention whether HTM's are a part of that technology. It would be interesting to see what companies Numenta has been working within in the 14 months since Subutai's talk.

2. The website also contains a basic description of its new learning algorithms. It is difficult not to notice how huge of a leap forward that Numenta views these algorithms as. In one place, Numenta states that the new algorithms are a "radical" improvement. In another place, it states that the new learning algorithms are "far superior" to the old ones. One thing that I wish the website contained was some experimental results showing these huge improvements. One thing that I found confusing was its description of prediction in the new algorithms. It described prediction as something flowing up the hierarchy. That seems different from prediction as described in the original HTM theory, which envisioned incoming data flowing up the hierarchy and predictions flowing down the hierarchy. In any event, it was an interesting read.

Tidbits on Numenta

Wow has it been a slow summer for HTM news. I have never seen a period of time where Numenta's employees have made so few public appearances. Since Hawkins' talk in March, I haven't seen any mention of any Numenta speeches, interviews, or papers of any kind. A few things of note:

-In its June newsletter, Numenta mentioned that it decided not to attempt an HTM workshop this year, meaning that the next generation algorithms will not be out this year. In their words, they decided not to push for an "interim" release this year, but to delay the workshop and release to 2011. "Interim" was an interesting choice of words, suggesting that a more fully featured product will be the result when NUPIC 2.0 does come out.

-Dileep George has a new blog on his website. It is called Mind Matter, and is located at the following link.

-I saw an interesting blog post regarding a robot called Nao that can apparently show and understand emotion. The author claims that the creators of the new robot software were using HTM software in the robot. I have not been able to verify that claim. The only mention of HTM in the context of a Nao robot that I found was an article by Ben Goertzel in which he describes using an HTM for low level perception in a Nao robot. That article specifically states that he hasn't implemented the idea yet, however. Find the link here.

-Finally, Tomaso Poggio, a Professor in the Department of Brain and Cognitive Sciences at MIT, and one of the creators of a biology-based hierarchical learning model known at HMAX, has created a software model that uses GPUs to greatly accelerate software that is designed to emulate the cortex, such as HTM or HMAX. Poggio claims that the software accelerates these biology-based models by an amazing 80-100 times. Poggio is listed as a technical advisor on Numenta's website, so hopefully they are aware of this, especially given the increased computational demands of NUPIC 2.0.