distillpub / post--gan-open-problems

Open Questions about Generative Adversarial Networks

Post -- Exploring Bayesian Optimization

Breaking Bayesian Optimization into small, sizable chunks.

To view the rendered version of the post, visit: https://distill.pub/2020/bayesian-optimization/

Authors

Apoorv Agnihotri and Nipun Batra (both IIT Gandhinagar)

Offline viewing

Open public/index.html in your browser.

NB - the citations may not appear correctly in the offline render

post--gan-open-problems's Issues

RevGAN not listed?

The first table demonstrating capabilities of different kind of generative models tells that GAN is irreversible, probably in general sense. But note that prior to this work (Ouderaa & Worrall, 2019) published reversible GAN (RevGAN) for image-to-image translation.

Please add this information somewhere in the text.

Add HYPE to open problem 5

Open problem 5 (How Should we Evaluate GANs and When Should we Use Them?) lists a number of proposed metrics for evaluating GANs. Zhou et al. recently (on April 1, 2019) proposed a new metric: HYPE: Human eYe Perceptual Evaluation of Generative Models (paper on arXiv; hype.stanford.edu), which is not yet in this list. Perhaps it'd be good to add it there?

I'm not sure what the process is for this, so instead of adding it myself and making a pull request, I figured it'd be easier to discuss it as an issue here first.