Processes Behind Reinforcement Learning: An examination of the temporal progression of processes behind successful and unsuccessful reinforcement learning strategies

The recent trend towards gamification could lead to an increase of the number of learning problems that need to be solved through reinforcement learning. It is therefore important that people learn how to solve reinforcement learning problems. Understanding which learning strategies people use, the processes behind them, and identifying sub-optimal learning strategies could prove very beneficial for teaching people the best reinforcement learning strategies. This study examines the processes behind reinforcement learning strategies through cognitive modeling. A reinforcement learning model was fitted on human behavior on a reinforcement learning problem. Some people were able to fully solve the problem and others were not. The temporal parameter trajectories of these two groups were compared to each other. The group that solved the problem showed expected results with a lot of learning and exploration at the start and less learning and more exploitation towards the end of the task....Show moreThe recent trend towards gamification could lead to an increase of the number of learning problems that need to be solved through reinforcement learning. It is therefore important that people learn how to solve reinforcement learning problems. Understanding which learning strategies people use, the processes behind them, and identifying sub-optimal learning strategies could prove very beneficial for teaching people the best reinforcement learning strategies. This study examines the processes behind reinforcement learning strategies through cognitive modeling. A reinforcement learning model was fitted on human behavior on a reinforcement learning problem. Some people were able to fully solve the problem and others were not. The temporal parameter trajectories of these two groups were compared to each other. The group that solved the problem showed expected results with a lot of learning and exploration at the start and less learning and more exploitation towards the end of the task. The other group started similarly but then started to learn less while exploration remained high. This could indicate that these people would benefit from short learning sessions after which they are able to focus on something else.Show less

Leiden University Student Repository

Documents

In Collections

Processes Behind Reinforcement Learning: An examination of the temporal progression of processes behind successful and unsuccessful reinforcement learning strategies