Quantum Computing has attracted increasing interest in recent years due to its potential for addressing problems that are difficult to solve with classical methods. Among the available technologies, Quantum Annealing provides a way to approach optimization problems by exploiting the physical evolution of a quantum system towards a low-energy state. While current hardware remains limited by noise and connectivity constraints, access to quantum annealers has become more widespread, enabling experimentation on practical tasks. Therefore, this thesis explores possible synergies between current Quantum Annealing and Machine Learning, in both directions. First, Quantum Annealing is applied to tasks commonly found in the Machine Learning field, such as Community Detection for recommendation and Feature Selection for recommendation, classification and ranking. These problems are reformulated for the quantum hardware, and the effectiveness of these approaches is evaluated through empirical studies. The formulation is versatile enough to represent both established and specifically designed problems. Quantum Annealing proves to be applicable to problems of around a hundred variables, requiring hybrid solutions for larger ones. Results are comparable or better than classical methods, but some drawbacks such as latency and problem programming time still remain. Then, Machine Learning methods are used to address limitations in the use of Quantum Annealing. A Reinforcement Learning model is developed to tackle the process of Minor Embedding, a necessary step for programming problems on current quantum annealers. Both feed-forward and graph-based neural network models are trained for this purpose, and their performance is analysed. Results show that a learning-based approach is feasible, and applying the right architectural and training choices also makes it more effective. The overall goal of this thesis is thus to evaluate the applicability of current Quantum Annealing hardware to Machine Learning problems and to explore how Machine Learning techniques can assist in improving Quantum Annealing workflows, paving the way for novel research and improvements on both fronts.
Il Quantum Computing ha suscitato un interesse crescente negli ultimi anni grazie al suo potenziale nel risolvere problemi difficili da affrontare con metodi classici. Tra le tecnologie disponibili, il Quantum Annealing offre un modo per affrontare problemi di ottimizzazione sfruttando l’evoluzione fisica di un sistema quantistico verso uno stato a bassa energia. Sebbene l’hardware attuale rimanga limitato dal rumore e dai vincoli di connettività, l’accesso ai quantum annealer è diventato più diffuso, permettendo di sperimentarli su problemi reali. Di conseguenza, questa tesi esplora possibili sinergie tra il Quantum Annealing attualmente esistente e il Machine Learning, in entrambe le direzioni. In primo luogo, il Quantum Annealing viene applicato a problemi tipici del Machine Learning, come la Community Detection per la raccomandazione e la Feature Selection per la raccomandazione, la classificazione e il ranking. Questi problemi vengono riformulati per l’hardware quantistico e l’efficacia degli approcci viene valutata tramite studi empirici. La formulazione è abbastanza versatile da rappresentare sia problemi esistenti sia problemi progettati appositamente. Il Quantum Annealing risulta applicabile a problemi di circa un centinaio di variabili, mentre per quelli più grandi sono necessarie soluzioni ibride. I risultati sono comparabili o migliori rispetto ai metodi classici, ma permangono alcuni svantaggi, come la latenza e il tempo di programmazione del problema. Successivamente, metodi di Machine Learning vengono usati per affrontare delle limitazioni nell’uso del Quantum Annealing. Viene sviluppato un modello di Reinforcement Learning per gestire il processo di Minor Embedding, un passaggio necessario per programmare i problemi sugli attuali quantum annealer. A questo scopo vengono addestrati sia modelli neurali feed-forward sia modelli basati su reti neurali su grafi, e ne vengono analizzate le prestazioni. I risultati mostrano che un approccio basato sull’apprendimento è fattibile e che, adottando le giuste scelte architetturali e di addestramento, può diventare anche più efficace. L’obiettivo complessivo della tesi è quindi valutare l’applicabilità dell’hardware attuale di Quantum Annealing a problemi di Machine Learning ed esplorare come le tecniche di Machine Learning possano contribuire a migliorare i flussi di lavoro del Quantum Annealing, aprendo la strada a nuove linee di ricerca e a miglioramenti su entrambi i fronti.
Developing the synergies between quantum annealing and machine learning
Nembrini, Riccardo
2025/2026
Abstract
Quantum Computing has attracted increasing interest in recent years due to its potential for addressing problems that are difficult to solve with classical methods. Among the available technologies, Quantum Annealing provides a way to approach optimization problems by exploiting the physical evolution of a quantum system towards a low-energy state. While current hardware remains limited by noise and connectivity constraints, access to quantum annealers has become more widespread, enabling experimentation on practical tasks. Therefore, this thesis explores possible synergies between current Quantum Annealing and Machine Learning, in both directions. First, Quantum Annealing is applied to tasks commonly found in the Machine Learning field, such as Community Detection for recommendation and Feature Selection for recommendation, classification and ranking. These problems are reformulated for the quantum hardware, and the effectiveness of these approaches is evaluated through empirical studies. The formulation is versatile enough to represent both established and specifically designed problems. Quantum Annealing proves to be applicable to problems of around a hundred variables, requiring hybrid solutions for larger ones. Results are comparable or better than classical methods, but some drawbacks such as latency and problem programming time still remain. Then, Machine Learning methods are used to address limitations in the use of Quantum Annealing. A Reinforcement Learning model is developed to tackle the process of Minor Embedding, a necessary step for programming problems on current quantum annealers. Both feed-forward and graph-based neural network models are trained for this purpose, and their performance is analysed. Results show that a learning-based approach is feasible, and applying the right architectural and training choices also makes it more effective. The overall goal of this thesis is thus to evaluate the applicability of current Quantum Annealing hardware to Machine Learning problems and to explore how Machine Learning techniques can assist in improving Quantum Annealing workflows, paving the way for novel research and improvements on both fronts.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Nembrini.pdf
accessibile in internet per tutti a partire dal 09/12/2026
Descrizione: PhD Thesis - Riccardo Nembrini
Dimensione
10.69 MB
Formato
Adobe PDF
|
10.69 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/248158