Dániel Varga

[Dániel Varga @danielxvarga]

@minchoi That TV screen at 4 seconds is where SOTA video generation was in 2020.

[Dániel Varga @danielxvarga]

Proof outline: cov f efe = cov efe efe > 0.

[Dániel Varga @danielxvarga]

"There's no such thing as a negative cov(F, E(F|E))." (remark by Peter Frenkel) #conditionalexpectation #positivesemidefiniteness #covfefe

[Soumith Chintala @soumithchintala]

Wasserstein GANs pretty aptly summarized in this reddit comment:
https://t.co/tOEvHG8l9f

[Dániel Varga @danielxvarga]

@alexjc Has anyone done t-SNE on style transfer Gram matrices? Map of Brushwork?

[Dániel Varga @danielxvarga]

@DeepSpiker I wonder how data-augmented training examples would fare.

[Dániel Varga @danielxvarga]

@fchollet To be Frank with you, François, I prefer 'dilated convolution'. But you have earned the right 2 name your classes however you wish

[Dániel Varga @danielxvarga]

@dribnet Please stay with the VAE! :) Seriously, though, WDYT, is this a question of tuning , or is GAN inherently inferior on this task?

[Dániel Varga @danielxvarga]

@fchollet I think any reasonable math6ian understands that lack of math. grounding is a plus: a huge research opportunity for math6ians.

[Dániel Varga @danielxvarga]

@iandanforth @brandondamos No, 20% was achieved earlier with standard wide resnet 1605.07146v1. (Our team is at 19% w/ wide stochd. resnet.)