Optimizing Neural Network Performance in Game Playing Using Simulated Annealing and Reinforcement Learning
Henrique Coutinho Layber, Vitor Berger Bonella, Flávio Miguel Varejão · 2024
This paper experiments optimization for Neural Network (NN) parameters for game playing using Simulated Annealing (SA) and Reinforcement Learning (RL). The study focuses on the Dino Game, comparing the performance of the proposed NN method against a baseline Decision Tree method. Experimental results demonstrate that the NN outperforms the Decision Tree, achieving a higher mean score with greater consistency. Statistical tests confirm the performance improvements are statistically significant, indicating the effectiveness of the SA heuristic in optimizing NN parameters.