Stanford CS236: Deep Generative Models I 2023 I Lecture 8 - GANs

For more information about Stanford's Artificial Intelligence programs visit: stanford.io/ai
To follow along with the course, visit the course website:
deepgenerativemodels.github.io/
Stefano Ermon
Associate Professor of Computer Science, Stanford University
cs.stanford.edu/~ermon/
Learn more about the online course and how to enroll: online.stanford.edu/courses/c...
To view all online courses and programs offered by Stanford, visit: online.stanford.edu/

[@CPTSMONSTER]

17:55? Log-likelihood derivation 21:50 Parameterize f theta or f inverse theta 28:50 Desiderata, Jacobians of special structure for fast determinant computation, lower triangular matrix determinant is product of diagonal elements 33:00? Derivation of Jacobian (simple layered coupling) 36:55 Simple layered coupling model is already able to map a complex distribution over pixels to a Gaussian (volume preserving) 42:00? Derivation of Jacobian (layered coupling with scaled shifts) 45:30? Generation process is deterministic 47:20 Sample data to get z, interpolation of z generates interpolated samples. Proves that while z are not compressive, they are meaningful latent representations. 50:00? Flow interpretation of autoregressive gaussian model 56:00? Inverse mapping computed in parallel 56:30? Jacobian lower diagonal because x_i only depends on z_

[@jesusmtz29]

That’s interesting. Is what happening in NICE that you’re almost mixing values by folding and adding.

[@chongsun7872]

This is the wrong title I think. It should be Flow models not GANs

[@legendarystuff6971]

First