Done for the class of Quantum Data Science, University of Minho
Masters in Physics Engineering - Information Physics Branch - 2022/23
We conducted a comprehensive investigation into the application of quantum reinforcement learning for solving the Cart Pole environment, comparing it with a classical model based on deep neural networks. Our study explored various quantum models, examining the impact of different entanglement layer configurations and the utilization of data re-uploading. Additionally, we varied the number of layers in the deep neural network.
Our findings indicate that, for models with fewer than four layers, the classical model demonstrates compatibility with the quantum model. However, as the number of layers increases, the quantum model outperforms the classical one. Remarkably, the quantum model exhibited the best performance during both training and testing stages.
Overall, our study highlights the advantages of employing quantum reinforcement learning for the Cart Pole environment, showcasing the superiority of certain quantum models over classical counterparts.