Reinforcement Learning

  Reinforcement learning is a framing of enabling agents to learn by trial and error from interaction with environments. Reinforcement learning has some problems, for example, the delay of the evaluation (reward) for agent’s action, the long learning time, and the trade-off between exploration and exploitation. We research methods for solving these problems.



Self-Organizing Map

  Self-organizing map (SOM) is one of neural networks with unsupervised competitive learning. SOM has superior performance of interpolation and robustness. By using these advantages, we research a supervised learning method for agent cooperative actions and a clustering method of software by SOM with user feedback.



Multi-agent Systems

  Multi-agent system is a system with interacting agents. We research a method of autonomous positioning of soccer agents based on Delaunay triangulation.