Deep Generative Models

 Today's post take a look at a great lecture by Ava Soleimany from the MIT Intro to Deep Learning course.  We'll be taking a look at Latent variable modesl, and discuss the differences between using variational autoencoders vs Generative Adversarial Networks (GAN) for creating generative models.

You can think of variational auto-encoders (VAE) as compression - decompression systems.  An image is compressed to a reduce representation that is the latent space in the first half of the VAE architecture.  The second half then reconstructs the inout image from the latent space representation.


GAN's are somewhat different, in that they are composed of a Generative part that takes an input vector (usually noise) and then generates an image from that. A second adversarial critic part then observes the output of the generator and makes a binary classification (is the generator output good or bad).  This result is then feed back to the complete system to force a competition where the Generator tries to create more realistic images, and the critic tries to get better at detecting real vs fake images created by the generator.


Here's a fun question to think about.  What is the latent space of a GAN?  We'll get into that more in future posts.


I've got another excellent video here for you to watch that does a very good job of breaking down how a Variational AutoEncode works.  Watching this will also give you a much better understanding of what the latent space is, and how you can work with it.  Dis-entangling the latent space is a big issue in getting GAN systems to be more controllable, so this information will be useful to better understand when we take our deep dive in to GAN systems later this week.


Note that both of these videos spend time explaining how to restructure VAE systems so that the probablistic part can be gradient optimized.  I caught that was important in my first view of Ava's presentation, but still didn't fully understand what they did, so having Xander explain it again is really useful to help drive it home.


We'll be continuing our 'deep' exploration of 'deep generative models' in thursday's 'deep dive into GANs' post

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