Game-based education with an AI learning companion

Ever since the remarkable achievement of AlphaGo by Google in 2017, it has led to the growing interest of reinforcement learning to be applied to modern-day solutions. One such application would be in educational serious games where reinforcement learning algorithms such as Q-Learning and Deep Q-Networks could be used to boost interactions between the game and players while helping them to learn and play the game better. In this paper, we chose Deep Q-Networks as our reinforcement learning algorithm which relies on artificial neural network for training of its model. We performed initial testing to develop an optimal neural network structure before training the model for the game. After the model has been trained, we evaluated the results of the model and integrated the model with the game for further testing. Our results have shown that the trained model is capable of guiding players through their mistakes to learn and play their game better. Recommendations for possible future extension to this project were also discussed.Bachelor of Engineering (Computer Science