iRobotLab

研究/各研究の話

Efficient learning by making a choice of sensor information†

Summary†

• Focusing of a relationships between sensor information and degrees of achievement of task.
• Calculating coefficient of correlation between sensor value and reward on reinforcement learning as degrees of achievement of task.
• Agent select important sensors for task based on coefficient of correlation.
• Agent reduces Q-space which is used important sensors only.

Conceptual diagram†