The Thousand Brains Theory

 Jeff Hawkins of Numenta has a new book out that tries to explain the results of Numenta's research into modeling how the cortex of the brain functions, and how intelligent behavior build off of that underlying structure.


The thousand brains theory of intelligence puts forward a novel theory of how the neocortex works.  It proposes that each cortical micro-column builds a model of an object, and they then collectively bubble out a consensus about what they are modeling actually means.  Other people have discussed micro-columns being a replicated neural computation element before, so other than de-emphasing hierarchy, i'm not sure there is really anything new here.

The real meat of the theory is the part that delves into grid cells in the brain.  Specifically grid cells in the neo-cortex.  And the notion that all computation in the cortex is ultimately composed of replicated grid cell structure that is used for all modality of sensory input, and for the mechanics of thought itself.

Grid cells are used by the brain to form a map of the environment.  Specifically, they represent the location of the organism in the environment, and how to navigate between that position to another one in the environment.  They also point out how reference frames are a fundamental element of thinking and intelligence.

I'm not really doing this theory justice in my brief overview above, so check out this more detailed blog post.  And i highly recommend reading the book.  It is written for layman, so no heavy technical jargon or math.

The google talk below does a good job of describing the theory from Hawkins and Subutai Ahmad. of Numenta.


Observations

1.  Replication of a single structure is a common thread in evolution, dna replication errors, and the genetics of development in multi-celled organisms.

2.  Relationship to Capsule Networks.

3.  Specific things discussed that could be applied to re-engineering deep learning systems

Sparse connections

  - benefits to sparsity

        minimal interference

        speed

        power consumption


neuron modeling

    continuous learning

    localized optimization (not global backprop)

    dendrites build the consensus between cortical columns via local massively distributed processing


grid cells

    reference frames

        object oriented

Comments

Post a Comment

Popular posts from this blog

CycleGAN: a GAN architecture for learning unpaired image to image transformations

Image
 Like the Doublemint Twins touting the joys of Doublemint gum, 2 GANs are surely better than 1 GAN.  Especially if we package them together inside of one meta GAN module.  And this is exactly what the CycleGAN architecture does. Have you ever harbored dark secrets of turning a horse into a zebra?  The CycleGAN was developed to do just that. Learn how to turn a horse into a zebra.  And more. Now right away you can notice a difference between the image to image transformation GAN architectures we've been discussing over the last few posts.  Those last few posts described systems that learn from a database of matched input-output image pairs. And if your goal is to turn an edge representation into a nicely filled in continuous tone image, it's easy to build your database of matched input-output image pairs that your GAN system can then learn off of.  Take a continuous tone image (which will be the output of the database pair entry), then run it through an edge detector algorithm. 
Read more

Pix2Pix: a GAN architecture for image to image transformation

Image
 I thought following up yesterday's TraVelGAN post with a Pix2Pix GAN post would be useful to compare what is going on in the 2 architectures.  Two different approaches to the same problem. I stole this Pix2Pix Overview slide below from an excellent deeplearning.ai GAN course (note that they borrowed it from the original paper) because it gives you a good feel for what is going on inside of the Pix2Pix architecture.    Note how the Generator part is very much like an auto-encoder architecture, but rebuilt using the U-Net architecture features (based on skip-connections) that fastai has been discussing in their courses for several years before it became more widely known to the deep learning community at large  (and which originally came from an obscure medical image segmentation paper) . So the Generator in this Pix2Pix GAN is really pretty sophisticated, consisting of a whole image to image auto-encoder network with U-Net skip connections to generate better image quality at highe
Read more

Smart Fabrics

Image
"What makes smart fabrics revolutionary is that they have the ability to do many things that traditional fabrics cannot, including communicate, transform, conduct energy and even grow."  This is a quote from Rebeccah Pailes-Friedman, who is a professor at Pratt with a research focus on developing wearable technology and smart textiles. Enhancing textiles can be viewed through different lens.  Enhancements can be aesthetic, or performance boosting, or durability boosting. 
Read more