Logo

MINDS & CIS Seminar - Anima Anandkumar - Shared screen with speaker view
Jeremias Sulam
20:13
Feel free to either post questions here in the chat, or alternatively raise your hand virtually and I'll call you to ask your question
Jason Eisner
49:33
Anima, now that you have a generative model for cats and non-cats that yields a CNN discriminator, can you train it on unlabeled data? Presumably that helps to train the z variables, but does it also improve the CNN?
Jason Eisner
55:27
Great, thanks Anima. Yes, small labeled sets + large unlabeled sets is where I expected this to pay off in practice, as you said.