Part 2 Lesson 12 wiki

(Caleb Everett) #77

Can GANs be used for data augmentation?


A way to detect naiive copying in GANs is to find similar images in the CNN feature space.
This was done by somebody on the famous nvidia progressive GAN (with faces).

found the link:

(Julio Barros) #79

They are actually pretty cool … as abstract art.

(Alex) #80

no, it gets closer to “jeremy’s” only after 75-100 epoch

(Mike Kunz ) #81

@emilmelnikov Mondrian

(Bart Fish) #82

Any GANS for music?

(Rachel Thomas) #83

Here is Helena’s twitter


A side note:

This is a paper that gets crazy good results (not really based on GANS) that I tried to implement.

My implementation (poorer quality results) :


RNNs works better for sequential data than GANS:

There is also Wavenets which also work great on sequential data generation.

(Erin Pangilinan) #86

@Interogativ See Project Magenta from Google


All of this sounds awfully like translating in one language to another then back to the original. Have GANs or any equivalent been tried in translation?

(Kevin Bird) #88

I was just thinking the same thing. It’s an interesting idea for sure.

(Sharwon Pius) #89

How to implement a GAN for a time-series model ? I’m not able to wrap my mind around it.


almost exactly what you mentioned:

Unsupervised Machine Translation Using Monolingual Corpora Only

(shameless plug)
my summary on that paper:



(Erin Pangilinan) #92

Can anyone share the link of the gif/interactive diagram or diagram Jeremy showed earlier (couldn’t find it?)

(chunduri) #93

please briefly explain vae, compared with gan’s?

(K Sreelakshmi) #94

(Kevin Bird) #95

That paper looks awesome. Thanks for sharing.

(Erin Pangilinan) #96

Thx I forgot to scroll lol.