Unifying VAEs and Flows into one Framework

 Most people break down generative deep learning models into one of 3 categories.  They are GANs, VAEs, and Flows.  We have covered the first 2 quite a bit here at HTC.  We have not really done so with the Flow architectures.

I've been trying to grok Flows recently, and came across this very interesting presentation by our old friend Max Welling called 'Make VAEs Great Again: Unifying VAEs and Flows'.  In it, he explains both, lays out the differences between them, an then tries to setup his own unified field theory of generative models where you can analyze them both in the same framework.

Yann LeCun has his own 'unified field theory' of generative models as well (energy based models), which we covered in a previous post.


Popular posts from this blog

Simulating the Universe with Machine Learning

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

Pix2Pix: a GAN architecture for image to image transformation