2. What weโll see today
โ Generative vs. Discriminative models [1]
โ VAE Algorithm Overview [2]
โ Putting it to work - Semi-supervised [3]
[1] Deep Neural Networks are Easily Fooled
[2] Auto-Encoding Variational Bayes
[3] Semi-Supervised Learning with Deep Generative Models
8. What do we want?
โ Generative model
โ โStructure constraintโ on latent space
Why?
โ Semi-Supervised learning
โ Visualize z-space
โ Not so easily fooled
โ More...
9.
10.
11. AutoEncoder Attempt #1
โ Encoder
q(z|x): get z given x
โ Decoder
p(x|z): get x given z
โ Whatโs the difference?
16. GAN - Generative Adversarial Networks
โYou pit a generative (G) machine against a discriminative (D) machine
and make them fight.โ
ยฉ Soumith Chintala
http://soumith.ch/eyescream/
21. If you do it right!
http://www.dpkingma.com/sgvb_mnist_demo/demo.html
22. Take Aways
โ Employ Structure, Itโs cool
โ GAN may be your next loss function
โ Super Res
โ Pixel Level Seg.
โ AutoEncoder (we saw it today)
โ Re-Parameterization Trick
----Personal takeaways-----
โ Donโt give up when it doesnโt work the first time (x1000)