21/22 – GANs

In these lectures, at long last, we will discuss Generative Adversarial Networks (GANs). GANs are a recent and very popular generative model paradigm. We will discuss the GAN formalism, some theory and practical considerations.


Reference: (* = you are responsible for this material)


7 thoughts on “21/22 – GANs

  1. I know it’s late to ask this (I think there are 5 courses left) and I assume that you (Aaron) must be incredibly busy, but would it be possible to post the course’s content 1 to 2 days prior to the lecture? In my case, it makes all the difference between a lecture in which the high level and abstract concepts are reviewed and I do not understand many concepts, so this often is not as helpful as it could be if I understood and reviewed the underlying details beforehand.

    In fact, if you could post way in advance the Deep Learning book sections to read on our own, not even the slides, this would be terrific and probably wouldn’t require too much time from you if you already know the planned topics and it would make a huge difference! Or also post external links to relevant slides and papers in advance too. Then you could edit your post later when the slides are ready.

    I mean, the course is great, but it often assumes prerequisites that some of us may not have in order to be able to follow a class with no preparation. It sometimes feels to me like a conference talk aimed at experts in the field.

    I don’t want to sound too negative, I love the structure of the course and it’s very exciting, the project, the blog, the questions. I just want to get clearer understanding out of it if possible!



    Liked by 2 people

    • Thank you, I appreciate this so much!

      Sorry if I was too harsh, I wrote this quickly in a moment of stress and frustration (at myself), please note that this was in no way a criticism or your teaching abilities, I’m sure most students would agree with me that you are a great pedagogue and an engaging teacher. It’s just a matter of some students that require preparation ahead of the course. For example, this is my first semester, so obviously I didn’t take the prerequisite for this course and my domain of expertise is narrow.


  2. I am not sure if people still look at the comments on the lectures but I was searching on mode collapse and GANs for the questions assignment and stumbled on the following paper:

    “AdaGAN: Boosting Generative Models” by I.Tolstikhin et al (2017)


    I thought it was an interesting idea and just wanted to share.



    • P.S: It does also make use of f-divergence and reference that paper as well so I thought it would make for a neat complement to the rest of the lecture and suggested reading


  3. One interesting application of GAN is to use GAN for Reinforcement learning. Here are two interesting papers, discussing actor critic and policy gradient with GAN.
    [1]Pfau D, Vinyals O. Connecting generative adversarial networks and actor-critic methods[J]. arXiv preprint arXiv:1610.01945, 2016.
    [2]Yu L, Zhang W, Wang J, et al. Seqgan: sequence generative adversarial nets with policy gradient[C]//Thirty-First AAAI Conference on Artificial Intelligence. 2017.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s