Aravind Srinivas: New paper - Reinforcement Learning with Augmented Data (RAD)! Data augmentation *alone* can achieve SoTA on DMControl and test-time generalization on ProcGen!
Project Page: https://mishalaskin.github.io/rad/
Code: https://github.com/MishaLaskin/rad https://t.co/WyvmeWmJY9
6 replies, 522 likes
Michael (Misha) Laskin: New updates to RAD (RL + data augs) answer the following:
1)Why does random crop work so well? -> translation invariance
2)Does data aug work for state-based RL too? -> yes
SOTA on DeepMind control (pixel-based RL) and OpenAI gym (state-based RL).
4 replies, 136 likes
Ekin Dogus Cubuk: Cool work that shows image augmentations can help RL generalization significantly. Also interesting to see 'crop' was the most effective operation in RL — I would say this is the case for vision tasks (both training and test time) too.
0 replies, 31 likes
Markus Wulfmeier 🏡: 'Nothing is as powerful as an idea whose time has come!'
Image augmentations are enabling some considerable performance boost in Deep RL
1 replies, 15 likes
Popular ML resources: The most popular ArXiv tweet in the last 24h: https://twitter.com/Aravind7694/status/1256026534130757632
0 replies, 3 likes
roadrunner01: Reinforcement Learning with Augmented Data
project page: https://mishalaskin.github.io/rad/ https://t.co/5hIrGswaeo
0 replies, 3 likes
Animesh Garg: Key Idea: Inferring local causal models (LCMs) from a global causal model given an object-oriented state representation. LCMs can generate counterfactual experiences that are causally valid in the global model.
HER (https://arxiv.org/abs/1707.01495) and RAD (https://arxiv.org/abs/2004.14990). https://t.co/LngnAz15EZ
1 replies, 3 likes
HotComputerScience: Most popular computer science paper of the day:
"Reinforcement Learning with Augmented Data"
0 replies, 1 likes
Found on May 01 2020 at https://arxiv.org/pdf/2004.14990.pdf