Communications of the ACM
UC Berkeley Researchers Open-Source RAD to Improve Any Reinforcement Learning Algorithm
Researchers at the University of California, Berkeley have open-sourced Reinforcement Learning with Augmented Data, a module that applies data augmentations to visual observations to achieve state-of-the-art results on common benchmarks.
Credit: Charlie Nguyen
A team of researchers at the University of California, Berkeley (UC Berkeley) has open-sourced Reinforcement Learning with Augmented Data (RAD).
The module applies data augmentations for visual observations to achieve state-of-the-art results on common benchmarks, and equals or outpaces every baseline in terms of performance and data efficiency across 15 DeepMind control environments.
In an accompanying paper, the team reports RAD can improve any existing reinforcement learning algorithm, and says it achieves better compute and data efficiency than Google AI's PlaNet, DeepMind's Dreamer, and UC Berkeley and DeepMind's SLAC.
"By using multiple augmented views of the same data point as input, CNNs (convolutional neural networks) are forced to learn consistencies in their internal representations," according to the paper.
From VentureBeat
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found