RL Generalization in a Theory of Mind Game Through a Sleep Metaphor (Student Abstract)

Training agents to learn efficiently in multi-agent environments can benefit from the explicit modelling of other agent's beliefs, especially complex limited-information games such as Hanabi card game. However, generalization is also highly relevant performance these games, though model comparisons at large training timescales be difficult. In this work, we address by introducing a novel trained using sleep metaphor on reduced complexity version This consists an altered regiment, well information-theoretic constraint policy. Results experimentation demonstrate improved through sleep-metaphor method, and provide promising motivation for similar techniques more methods that incorporate models beliefs.