From Quantum Quirks to Clear Skies: A Not-So-Dusty Road to Sensor Calibration

This thesis investigates the application of Quantum Machine Learning (QML) and Deep Learning to the calibration of optical fine-dust sensors, aiming to enhance air quality monitoring within the smart city framework, specifically in Turin. The calibration of these sensors is crucial for deploying a city-wide sensor network that can accurately and efficiently monitor air pollution levels, a step forward in the evolution of smart cities. By leveraging the potential of QML, this work seeks to address the limitations of current sensor calibration methods, thus contributing to the development of more resilient, efficient, and accurate environmental monitoring systems. Smart cities represent the next frontier in urban management, promising enhanced sustainability, livability, and resource efficiency through technology. Central to this vision is the capability to monitor and manage environmental parameters accurately, wherein air quality plays a critical role. The calibration of optical fine-dust sensors using advanced machine learning techniques is pivotal for this purpose, ensuring that data collected is reliable and actionable. The thesis is structured to first provide a comprehensive background on smart cities, the importance of air quality monitoring, and the challenges faced in sensor calibration. It then introduces the linear regression problem as the foundation for calibration, discussing various optimizers, loss functions, and deep learning models, with a focus on RNNs and LSTMs due to their efficacy in handling time-series data. The advent of quantum computing offers a novel approach through Variational Quantum Circuits, promising to revolutionize sensor calibration by harnessing quantum mechanics' computational advantages. A detailed comparison between classical and quantum machine learning models forms the core of the research. Through rigorous testing, including hyperparameter tuning and cross-validation, the thesis evaluates the performance of traditional neural networks against quantum models in sensor calibration. Despite the nascent stage of quantum hardware and the current superiority of classical models in terms of accuracy, the findings highlight the promising future of quantum models in computational efficiency and potential scalability. The conclusion underscores the significance of advancing QML for environmental monitoring within smart cities. Although Quantum LSTM models are still under development, their evolving capabilities indicate a promising direction for achieving enhanced sensor calibration. This research not only contributes to the body of knowledge in QML and environmental monitoring but also outlines a future where smart cities can leverage quantum computing to improve quality of life, environmental sustainability, and urban management. In essence, this thesis underscores the importance of innovating calibration techniques for environmental sensors in smart cities through quantum computing. It highlights the current challenges, the comparative analysis of classical and quantum approaches, and the future potential of QML to revolutionize air quality monitoring, making a significant contribution to the smart city paradigm and the broader field of IoT sensors aiding human society.