EvoJAX is a hardware-accelerated neuroevolution toolkit built on top of JAX. It can help run a wide range of evolution experiments within minutes on a TPU/GPU, compared to hours or days on CPU clusters.
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.
Agents with a self-attention “bottleneck” not only can solve these tasks from pixel inputs with only 4000 parameters, but they are also better at generalization.
