Permutation-Invariant Neural Networks for Reinforcement Learning

Reinforcement learning agents typically perform poorly if provided with inputs that were not clearly defined in training. A new approach enables RL agents to perform well, even when subject to corrupt, incomplete, or shuffled inputs. Note: This blog post about our paper is written by Yujin Tang and myself, and was originally posted on Google AI Blog. It has been cross-posted here for archival purposes. Introduction “The brain is able to use information coming from the skin as if it were coming f...