Fumito Uwano

Student in the doctoral program of the University of Electro-Communications from April 2017 to March 2020 (and Research fellowship DC1 for young scientists in Japan Society for the Promotion of Science)

Uncommunicative Multi-Agent Reinforcement Learning on Static and Dynamic Environments
I proposed the method enables agents to learn cooperative policies without any communications, as well as extended it to dynamic environments in which some accident and others are happened to achieve agents' cooperation under the situation communication delay and it's incorrect information as the real world problems. The results to apply the proposed method to maze problems derived that it can get the maximum rewards as the fastest than Q-learning.

Assistant Professor in Okayama University from April 2020 to present

Multi-Agent Reinforcement Learning to Acquire Cooperative Policy with Different Abstraction
I propose the cooperative method to solve the hetero informatic problem by which agents with different resolutions. Concretely, the proposed method designs agents to control the inputs' abstractions as well as manage the abstractions along to their resolutions. It makes them learn cooperative policies in that problem.

Transfer Learning for Multi-Agent System to Adapt to Unknown Cooperation and Environment
I propose the method for agents to learn cooperative policy on unknown environment, as well as unknown cooperative policy. Concretely, the proposed method designs new primitive cooperative policies based on learned knowledges to conbine them and learn on the combined policies. it makes them create unknown cooperative policies and learn cooperative policies on the unknown environment.