Introduction to GANs

Gudmundur presentation

I will upload a summary of the presentation, along with the presentation itself after the journal club meeting, which is scheduled for the 3rd of October.

For now, those who are eager to get a headstart on GANs, I recommend reading the original paper by Ian Goodfellow et al. This original paper from 2014 has currently been cited almost 5000 times. I will also try to find some more up to date material, I will progressively update this post with papers and blog posts relevant for the presentation. GANs also get a pretty good treatment in the Deep Learning book. For an overview of all GANs with a special name, check this GAN zoo.

Update: The presentation can be found here.

In the presentation I went into the basics of GANs and what they can be used for. There are several avenues that we can use these, specifically in relation to synthesizing training data. In the coming weeks we shall focus more on the recent advances in GANs, like the cycle loss and tricks to generate more high resolution images.