As quantum hardware progresses through the NISQ era, new software solutions are proposed, enhancing the qualities of both the quantum and the classical setting. Increasingly complex tasks prove to be viable through these new architectures, making Machine Learning (ML) algorithms benefit from quantum computing and allowing for the development of Quantum Machine Learning (QML) models. Our thesis focuses on Variational Quantum Classifiers (VQC), a hybrid quantum-classical architecture that combines variational quantum circuits with classical optimization methods used for ML. Among this family of QML models, we study how they behave for two-local architectures, which are circuits composed by alternating layers of rotation blocks and entanglement blocks. As the choices made for the design of these circuits define the model itself, we explore this architecture along four dimensions: the rotation types in each layer, the entanglement gates in each layer, the topology of the entangling blocks and the number of layers of the model. Our objective is to assess how the components of the variational quantum circuit influence a TwoLocal VQC, how its depth influences the classification performance and whether it is possible to formulate experimental heuristics to navigate this wide range of possible combinations. The results in our thesis display the answers to those questions, as it analyzes over three different datasets the 60 possible combinations for the choices of our hyper-parameters: - rotation type: RX, RY, RZ, RYRZ, RZRYRZ - entanglement topology: chain, ring, full, pairwise - entanglement gate: CNOT, CZ, IsingXX Through this study, we are able to formulate practical guidelines for designing TwoLocal VQCs across multiple dimensions, problem complexity and number of layers, as we help the reader reduce experimental costs by focusing resources on the most impactful solutions.
Con il progresso dell'hardware quantistico nell'era NISQ, nuove proposte per soluzioni software capaci di sfruttare sia le qualità dell'ambiente quantistico che di quello classico vengono presentate. Grazie a queste nuove architetture, compiti sempre più complessi risultano realizzabili, consentendo agli algoritmi di Machine Learning (ML) di trarre vantaggio dal quantum computing attraverso lo sviluppo di modelli di Quantum Machine Learning (QML). La nostra tesi si concentra sui Variational Quantum Classifiers (VQC), un'architettura ibrida quantistico-classica che combina circuiti quantistici variazionali con metodi di ottimizzazione classici utilizzati per il ML. All'interno di questa famiglia di modelli QML, studiamo nello specifico le architetture a two-local, che sono circuiti composti da layer alternati di gate di rotazione e di entanglement. Poiché le scelte effettuate per la progettazione di questi circuiti definiscono il modello stesso, siamo in grado di esplorare questa architettura lungo quattro dimensioni: i tipi di rotazione in ogni layer, i gate di entanglement in ogni layer, la topologia dei blocchi di entanglement e il numero di layer del modello. Il nostro obiettivo è valutare come i componenti del circuito quantistico variazionale influenzano un VQC TwoLocal, come la sua profondità impatta sulle prestazioni di classificazione e se sia possibile formulare delle euristiche per navigare in questa vasta gamma di possibili combinazioni . I risultati della nostra tesi mostrano le risposte a queste domande, analizzando su tre diversi set di dati le 60 combinazioni possibili per le scelte dei nostri iperparametri: - tipo di rotazione: RX, RY, RZ, RYRZ, RZRYRZ - topologia di entanglement: ring, chain, full, pairwise - gate di entanglement: CNOT, CZ, IsingXX Attraverso questo studio, siamo in grado di formulare linee guida pratiche per la progettazione di VQC TwoLocal su più dimensioni, complessità del problema e numero di livelli, aiutando il lettore a ridurre i costi sperimentali concentrando le risorse sulle soluzioni più efficaci.
Experimental analysis of two-local quantum machine learning models
CIARALLO, ANDREA
2024/2025
Abstract
As quantum hardware progresses through the NISQ era, new software solutions are proposed, enhancing the qualities of both the quantum and the classical setting. Increasingly complex tasks prove to be viable through these new architectures, making Machine Learning (ML) algorithms benefit from quantum computing and allowing for the development of Quantum Machine Learning (QML) models. Our thesis focuses on Variational Quantum Classifiers (VQC), a hybrid quantum-classical architecture that combines variational quantum circuits with classical optimization methods used for ML. Among this family of QML models, we study how they behave for two-local architectures, which are circuits composed by alternating layers of rotation blocks and entanglement blocks. As the choices made for the design of these circuits define the model itself, we explore this architecture along four dimensions: the rotation types in each layer, the entanglement gates in each layer, the topology of the entangling blocks and the number of layers of the model. Our objective is to assess how the components of the variational quantum circuit influence a TwoLocal VQC, how its depth influences the classification performance and whether it is possible to formulate experimental heuristics to navigate this wide range of possible combinations. The results in our thesis display the answers to those questions, as it analyzes over three different datasets the 60 possible combinations for the choices of our hyper-parameters: - rotation type: RX, RY, RZ, RYRZ, RZRYRZ - entanglement topology: chain, ring, full, pairwise - entanglement gate: CNOT, CZ, IsingXX Through this study, we are able to formulate practical guidelines for designing TwoLocal VQCs across multiple dimensions, problem complexity and number of layers, as we help the reader reduce experimental costs by focusing resources on the most impactful solutions.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Ciarallo.pdf
accessibile in internet per tutti
Dimensione
4.13 MB
Formato
Adobe PDF
|
4.13 MB | Adobe PDF | Visualizza/Apri |
|
2025_12_Ciarallo_executive_summary.pdf
accessibile in internet per tutti
Dimensione
405.83 kB
Formato
Adobe PDF
|
405.83 kB | 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/10589/247483