Learning to Reinforcement Learn with Transformer [Manuscript in preparation]
We developed a transformer-based agent that achieves meta-reinforcement learning through in-context learning, demonstrating one-shot learning in a 4x4 recurrent spatial navigation task.