A CBR-ANN hybrid for dynamic environments

Abstract

This paper proposes a Case-based Reasoning (CBR) and Artificial Neural Network (ANN) hybrid solution for dynamic problems. In this solution, a CBR system chooses between several expert neural networks for a given case/problem. These neural networks are Recurrent Neural Networks, which are trained using Deep Q-Learning (DQN). The system was tested on the game Mega Man 2 for the NES, and is compared to how a single recurrent neural network performed. The results collected outperforms the basic ANN that it was compared against, and provides a good base for future research on the model.