Investigating Action Encodings in Recurrent Neural Networks in Reinforcement Learning

In three linesInvestigation of how action information is incorporated into RNN state update functions for reinforcement learning. Authors empirically evaluate several recurrent cell architectures on illustrative domains and identify critical design choices often unreported in large-scale RL agents.

Summary generated by Claude — human-verified