MLSS-Indo-GNN-Zoom
From Max Welling to Everyone: (5:57 PM)
Where shall I start today? Maybe finish my lecture from yesterday first?
From SAID AL FARABY to Everyone: (5:57 PM)
I think so
From Max Welling to Everyone: (5:57 PM)
Where did we end up yesterday?
From SAID AL FARABY to Everyone: (5:58 PM)
wait i’ll check the slide Slide 30 Reparameterization trick
From Max Welling to Everyone: (6:00 PM)
ok!
From Me to Everyone: (6:00 PM)
where can we get the slide?
From Max Welling to Everyone: (6:00 PM)
I will share afterwards with Said and he can put it on the website
From Me to Everyone: (6:01 PM)
ok! thank you very much :)
From chris simon to Everyone: (6:10 PM)
Can we ask questions here? on zoom I mean
From SAID AL FARABY to Everyone: (6:10 PM)
yes
From chris simon to Everyone: (6:10 PM)
cool
From Me to Everyone: (6:12 PM)
is the | | an absolute?
From chris simon to Everyone: (6:20 PM)
From your surjective diagram, z is not unique. Does it yield a bigger chance for posterior collapse? gen
From Renny P. Kusumawardani to Everyone: (6:24 PM)
For the stochastic process, it is necessary to take samples. I’ve heard the term ‘sample-efficiency’. What does it mean and why does it matter?
From Me to Everyone: (6:26 PM)
At log | det gradient_x(z) |, the notation | | means an absolute, right?
From SAID AL FARABY to Everyone: (6:31 PM)
Sorry, I missed it. We think so: https://deepgenerativemodels.github.io/notes/flow/ you can rise your hand if you want to talk directly
From Me to Everyone: (6:40 PM)
thank you very much for your confirmation :)
From Yopi to Everyone: (6:47 PM)
I might have missed, what is the role of the state memory in the hybrid method?
From chris simon to Everyone: (6:47 PM)
Most of your recent works focus on the explicit functions for sample generation such as VAE. Is it because sample generation with implicit functions (e.g. Langevin dynamics+ energy based models) not working really well in practice? What is your take?
From Arian Prabowo to Everyone: (6:51 PM)
I see thank you very much!
From Wawan Cenggoro to Everyone: (6:56 PM)
The neural augmentation remind me your work titled "Semi-Supervised Learning with Deep Generative Models", which is a two layer VAE with distribution of y at the last VAE layer. Do you think combining it can be beneficial? combining them I mean
From chris simon to Everyone: (6:57 PM)
I see thanks!
From Wawan Cenggoro to Everyone: (6:57 PM)
yes
From Ebuka Oguchi to Everyone: (6:58 PM)
Are there any measures to ensure these deep fakes images are not used for the wrong purposes?
From Wahyu Hidayat to Everyone: (7:23 PM)
wonderful
From chris simon to Everyone: (7:29 PM)
Wonderful!
From Renny P. Kusumawardani to Everyone: (7:39 PM)
Great advice!
From Arian Prabowo to Everyone: (7:47 PM)
That's a very good advise! thank you!
From Me to Everyone: (7:50 PM)
thank you so much !
From Robby Hardi to Everyone: (7:50 PM)
Great talk!