HTC Seminar #27 - Priors for System 2 Knowledge Representaation

 Today's HTC Seminar is by Yoshua Bengio on 'Priors for System 2 Knowledge Representaation', and was presented at the ICML conference in July 2020.

The talk is on Slides Live and can be seen here.

Observations

1.  So what is system 1 and system 2 (not to be confused with software 1.0 and software 2.0 metaphors)? If you ask the all knowing entity 'the googler', you get the following (at least you do today):

System 1 is the brain's automatic, intuitive, and unconscious thinking mode. It requires little energy or attention, but it is often biased prone. ... System 2 is a slow, controlled, and analytical method of thinking where reason dominates. Unlike system 1, it requires energy and attention to think through all the choices.

So Yoshua is taking this concept of cognition described by Daniel Kahneman in his book 'Thinking Fast and Slow', and applying it to deep learning. So system 1 is learning the data manifold, system 2 is more like traditional rule based AI.

Here's a quick look into a conversation with Daniel Kahnman if you are interesting in understanding more about this concept.



2.  So what are priors?

No, it's not refering to your previous criminal record.

If you are a Bayesian, then a prior is a probability distribution over a set of distributions which expresses a belief in the probability that some distribution is the distribution generating the data.

A broader way to think about i in the context of deep learning is that it's the 'knowledge' encoded within the structure of your architecture. So if you are doing data augmentation, you are pre-building priors into the architecture based on your human knowledge of what data augmentations seem perceptually relevant.

If you are using a pre-trained neural network in your system in some way, that is a prior in your system.

What is fascinating is that even if you use a randomly-initialized neural network, it can be considered a had-crafted prior that can lead to reasonable results on some problems.

Check out this paper for an example of what i mean by this.

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