Ian Goodfellow, Research Scientist OpenAI : Generative Adversarial Networks (GANs) #AIWTB 2016

Ian Goodfellow, Research Scientist at OpenAI takes us through Generative Adversarial Networks at AI With The Best online developer conference September 2016 #AIWTB.

Followed by a super Q&A session

Q&A
28:40 Can you explain the problem about GANs not knowing how to place body parts and textures within the image net - what are some solutions to those problems?
30:45 Specifically how can you place body parts and place 3D texture?
31:30 What are the commercial applications of GANs?
33:04 Where do you see GANs being applied outside of image related tasks?
33:13 How can you generate feedback loops in a GAN so it generates an output similar to the input in its dimensions to predict the next step in a time series ie. a video sequence like in the case of WaveNet?
33:34 What do you predict it will take to get the output to scale to a much larger output dimension?
34:16 Can GANs be used for Data Compression?
35:20 What is your goal with GANs?

[machine-learning]

Hey, saw you in an interview video with Siraj Raval where you very nicely introduced the concept of adversiarial networks. Good stuff, thanks for sharing!

[TheAIEpiphany]

Super useful presentation! Thanks for sharing!

[muhammadharris]

Great video on GANs by non other than the creator himself. though his voice a bit mesmerizing in a sense i feel sleepy going though the video. Anyways, a tremendous effort, hopefully we will see some solid improvements in terms of mode collapse and speedier training in th near future.

[hengo_tt]

Can we use GAN for 2 classes which cannot be classified clearly?

[adithyareddynallabolu]

Thanks, it is a good talk on GAN's

[Champignon]

Could using intermediate layers of discriminator for perceptual loss be an idea?

[shafighshahbazi]

Is CNN a requirement for implementing gan in practice or not?