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.


Redirecting to attentionagent.github.io, where the article resides.