Papers of the day   All papers

Task-Relevant Adversarial Imitation Learning

Comments

Oct 03 2019 Ali Eslami

Now for something different! Deep RL + GAN training + CelebA = artificial caricature. Agents learn to draw simplified (artistic?) portraits via trial and error. @ #NeurIPS2019 creativity workshop. Animated paper: https://learning-to-paint.github.io PDF: https://arxiv.org/abs/1910.01007 Thread. https://t.co/eeChwyP57f
4 replies, 618 likes


Oct 03 2019 Konrad Żołna

I am fortunate to have interned @DeepMindAI where we extended GAIL to make it work better for robot manipulation tasks. Task-Relevant Adversarial Imitation Learning (TRAIL) robustly learns policies and rewards from pixels Paper https://arxiv.org/abs/1910.01077 Demo https://youtu.be/46rSpBY5p4E https://t.co/R4RgNQiFPd
2 replies, 222 likes


Oct 06 2019 Nando de Freitas

Two recent works @DeepMindAI on generating embodied behaviour with GANs. https://arxiv.org/abs/1910.01077 and https://arxiv.org/abs/1910.01007
3 replies, 178 likes


Oct 03 2019 Nando de Freitas

The GAN / GAIL idea has started to work for robust off-policy reinforcement learning from pixels. This is now exciting and promising! TRAIL by @konradzolna @scott_e_reed @ziyuwang @SashaVNovikov @serkancabi @notmisha @davidmbudden and Sergio Gomez.
0 replies, 131 likes


Oct 03 2019 Scott Reed

New paper w/ @konradzolna: Task-Relevant Adversarial Imitation Learning TL;DR prevent discriminator from distinguishing agent/expert using task-irrelevant information, such as arm appearance or initial prop positions. Paper https://arxiv.org/abs/1910.01077 Demo https://youtu.be/46rSpBY5p4E https://t.co/oPlaSe6opn
1 replies, 15 likes


Oct 03 2019 ziyu wang

Our new paper on improving generative adversarial imitation learning with @konradzolna, @scott_e_reed, @SashaVNovikov, and @NandoDF.
0 replies, 10 likes


Oct 08 2019 Mike de Waal

Task-Relevant Adversarial Imitation Learning #ArtificialIntelligence #Task #ai via http://twinybots.ch https://arxiv.org/abs/1910.01077
0 replies, 2 likes


Content