Re-Inforcement Learning In Multi Agent System With Transfer Learning

Transfer learning is the process that uses the previous task knowledge in the current task to improve the performance of the new task. In reinforcement learning, the agent requires training from the task. Transfer learning use with the reinforcement learning to improve the performance of agent. Transfer learning method is mainly applied on single agent reinforcement learning algorithms. We use the better algorithm for transfer learning with reinforcement learning on multi agent domain. Domain is a real time strategy game (RTS). In multi agent one agent is a cooperative agent and another is competitive agent. In reinforcement learning the agent give limited feedback or information to agent. The main role of transfer learning is that the learning in one task can help to improve performance in another task. In multi agent framework the advice given by agent in one stage can be used in another stage of the system. Our experiment conduct on real time strategy game setup. The result shows that Bias Transfer reduces the training time in the target task and improves asymptotic performance.