Learning device that performs reinforcement learning of a policy of an agent, learning method, and computer-readable storage medium

Abstract

A learning device is configured to perform reinforcement learning of a policy of an agent by self-play under a multi-agent environment. The multi-agent environment is an asymmetric environment in which at least one of a type of an action performed by the agent, a type of a state acquired by the agent, and a definition of a reward given to the agent is different, and an evaluation index for learning is given to each of agents. The learning device performs performing learning of a predetermined agent of the agents by using a learning model, acquiring the evaluation index for the learning model of the predetermined agent after learning, comparing the evaluation index for the predetermined agent with the evaluation index for the other agent, and a step of setting the agent having a lower evaluation index as a learning target.