00:17:55	Antonio Torralba:	https://www.amazon.com/Numerical-Linear-Algebra-Lloyd-Trefethen/dp/0898713617/ref=sr_1_1?dchild=1&keywords=david+bau&qid=1629824109&sr=8-1
00:18:36	Terran Melconian:	I thought the explanation of SVD in the book was very clear and accessible. :)
00:18:50	Joshua Fasching:	Agreed
00:29:10	Henry A Lieberman:	Can you ask, “I want a green bed but white walls”? Or “green bed but not green walls”?
00:45:21	Rob Frost:	What if you remove 20 random neurons? Does removing the most 20 most important neurons reduce performance more than removing 20 random ones?
01:16:56	James Gilles:	Some researchers (Ilyas et al, "Adversarial examples are not bugs, they are features") have conjectured that neural networks can find predictive but non-human-perceptible features in images. Have you seen any evidence for/against this during your research? For instance, classifier neurons that are important to a class but aren't aligned with human-recognizable concepts?
01:28:59	Anna Ivanova:	great work!!