We were unable to load Disqus. If you are a moderator please see our troubleshooting guide.

Shashank Gupta • 4 years ago

Great Article! Thanks!

You seem to have a typo in your JS_{\pi} definition .. the first part should be multiplied with $(1 - \pi)$, and the second with $\pi$, instead of the other way round.

Jae Duk Seo • 2 years ago

thanks

Chris Inskip • 3 years ago

Great set of articles!

Just a quick comment, it seems P and Q are reversed in the parameterised JS divergence; am I mistaken?

Junhyun Cho • 4 years ago

Edward Choi • 4 years ago

Thank you for the nice article! I learned a lot about the nature of GAN from this.
BTW I think there is a typo in JSD[P||Q] = JSD[P||Q]. Maybe you meant JSD[P||Q] = JSD[Q||P]?

Yaodong Yang • 5 years ago

Hi Ferenc:

Thanks a lot for the articulated blog.

I was wondering if it is possible to share the code you had used to make the Figure 1? I assume it is an updated version of what Theis has done in "A note on the evaluation of generative models"?

Many thanks!

Hey, your link http://bethgelab.org/media/... does not work anymore. Could you tell me the name of the paper? Also, some of your equations do not render properly for some reason.

Ferenc Huszar • 5 years ago

Ah, thanks for reminding, that link's been broken for ages (an extra _ at the end of the filename). Here it is: http://bethgelab.org/media/...
And here's a link to arxiv: https://arxiv.org/abs/1511....