Date

2019

Author

Hasanoğlu, Mert.

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

186
views

65
downloads

Cite This

This study aims to find out the effect of prior knowledge on the success of humans in a non-rewarding game environment, and then to compare human performance with a reinforcement learning method in an effort to observe to what extent this method can be brought closer to human behavior and performance with the data obtained. For this purpose, different versions of a simple 2D game were used, and data were collected from 32 participants. At the end of the experiment, it is concluded that prior knowledge, such as the meaning of objects and colors, have an impact on human performance. It was observed that the reinforcement learning agent failed to finish the same game. Various attempts have been made to improve performance and to achieve human-like behavior. In one of these, mini-games were prepared to intro- duce prior knowledge of the objects and the interaction with them. In another trial, a model is created with the game data collected from participants, and the agent is trained using this model as an exploration strategy. Only when the human data is used as an exploration strategy, the agent succeeded in finishing the game. Although the performance of the reinforcement algorithm is increased, human-like behavior is not observed. The conclusion is that it is more meaningful to consider prior knowledge within the context of exploration strategy, and having prior knowledge is not enough in achieving human-like behavior.

Subject Keywords

Reinforcement learning., reinforcement learning, human priors, artificial neural networks, machine learning

URI

http://etd.lib.metu.edu.tr/upload/12625166/index.pdf
https://hdl.handle.net/11511/45543

Collections

Graduate School of Informatics, Thesis