A parameter control method inspired from neuromodulators in reinforcement learning

摘要

The brain gains appropriate behaviors which gets rewards and escapes punishments by trial and error. Reinforcement learning models such a nature's system by an engineering approach. Neuromodulators, which projects widely in brain and adjusts functions in each brain part, are matched with internal parameters of reinforcement learning. We propose a reinforcement learning algorithm which can follow sudden changes in environment by considering how neuromodulators affect behaviors. This method improves actions by controlling the internal parameters of reinforcement learning after the obtained reward decreased as compared with the past. We actually applied this algorithm to learning problems, with the result that it followed sudden changes in environment.