Memoroids Framework for Efficient Recurrent Reinforcement Learning

The memoroids framework reformulates recurrent models as monoids to improve batching in recurrent reinforcement learning. This approach addresses scaling issues in long sequences for Linear Recurrent Models and improves sample efficiency and returns compared to traditional RNN or Transformer mappings.