Anomaly Detection (AD) has gained significant attention due to its critical role in various high-stakes domains, including finance, cybersecurity, and healthcare. Classical AD techniques have been widely used, however, as data volume continue to grow, they face scalability challenges, calling for quantum-enhanced approaches to potentially achieve speed-ups. The fault-tolerant regime of quantum computing enables the error-free execution of complex algorithms at scale. While prior research in this regime has proposed high-level quantum algorithms for AD, most lack circuit-level implementations suitable for execution on quantum hardware and simulators, limiting their practical viability. This motivates the design of quantum circuits and frameworks to assess their advantages in practice. This thesis presents a complete quantum circuit design for an AD based on Quantum Principal Component Analysis (AD-QPCA) algorithm. The proposed method integrates Quantum Random Access Memory (QRAM) for superposition-based access to the data matrix and quantum Amplitude Estimation for efficient inner-product computation. Qiskit-based realization is also provided to validate the circuital designs and for future deployment. Unlike prior high-level formulations, this work decomposes the algorithm into the Clifford+T universal gate set, enabling precise resource accounting in terms of logical qubits, gate counts, and circuit depths. We present a comprehensive resource analysis that shows a logarithmic depth scaling with respect to the dataset size, confirming the exponential speed-up obtained in query complexity. We estimate the resources to apply the method to a real-world 2024 bank transaction dataset for fraud detection. Estimates reveal that current hardware limitations—particularly the absence of scalable, low-overhead QRAM devices—dominate the cost, requiring O(10^10) logical qubits for utility-scale deployment. This work provides concrete circuit-level designs for AD establishes concrete and identifies key bottlenecks, identifying concrete hardware performance targets for a quantum advantage in real-world AD tasks.
Il rilevamento delle anomalie (AD) ha acquisito crescente rilevanza in numerosi ambiti critici, tra cui finanza, cybersicurezza e sanità. Tecniche classiche per l’AD sono ampiamente impiegate, tuttavia, con l’aumento costante del volume dei dati, esse soffrono limitata scalabilità, motivando l’adozione di approcci quantistici per ottenere eventuali speed-ups. Il calcolo quantistico tollerante agli errori consente l’esecuzione di algoritmi complessi su larga scala. Sebbene studi in questo regime abbiano proposto algoritmi quantistici per l’AD a livello teorico, la maggior parte non fornisce implementazioni circuitali per l’esecuzione, limitandone così la fattibilità pratica. Ciò motiva la progettazione di circuiti e framework quantistici volti a valutarne i vantaggi pratici. Questa tesi presenta il disegno circuitale completo per un algoritmo di AD basato sull’Analisi delle Componenti Principali Quantistica (AD-QPCA). Il metodo proposto integra la Quantum Random Access Memory (QRAM) per l’accesso in sovrapposizione alla matrice di input e la procedura quantistica di Amplitude Estimation per un calcolo efficiente dei prodotti scalari. È inoltre fornita un’implementazione Qiskit per validare il disegno circuitale e favorire sviluppi futuri. A differenza di precedenti studi teorici, il presente lavoro scompone l’algoritmo nel set universale di porte Clifford+T, consentendo l’analisi precisa delle risorse in termini di qubit logici, numero di porte e profondità dei circuiti. L’analisi dimostra una scalabilità logaritmica della profondità rispetto alla dimensione del dataset, confermando lo speed-up esponenziale ottenuto nell’analisi in query complexity. Si stimano le risorse necessarie per applicare il metodo ad un dataset di transazioni bancarie del 2024 per la rilevazione di frodi. Le stime evidenziano che le attuali limitazioni hardware – in particolare l’assenza di dispositivi QRAM efficienti – costituiscono il fattore dominante, richiedendo O(10^10) qubit logici per l’impiego su scala. Questo lavoro fornisce circuiti concreti per l’AD e individua le principali limitazioni, definendo obiettivi di performance hardware necessarie a conseguire un vantaggio quantistico in pratica.
Quantum principal component analysis for financial fraud detection
Guerrini, Alberto
2024/2025
Abstract
Anomaly Detection (AD) has gained significant attention due to its critical role in various high-stakes domains, including finance, cybersecurity, and healthcare. Classical AD techniques have been widely used, however, as data volume continue to grow, they face scalability challenges, calling for quantum-enhanced approaches to potentially achieve speed-ups. The fault-tolerant regime of quantum computing enables the error-free execution of complex algorithms at scale. While prior research in this regime has proposed high-level quantum algorithms for AD, most lack circuit-level implementations suitable for execution on quantum hardware and simulators, limiting their practical viability. This motivates the design of quantum circuits and frameworks to assess their advantages in practice. This thesis presents a complete quantum circuit design for an AD based on Quantum Principal Component Analysis (AD-QPCA) algorithm. The proposed method integrates Quantum Random Access Memory (QRAM) for superposition-based access to the data matrix and quantum Amplitude Estimation for efficient inner-product computation. Qiskit-based realization is also provided to validate the circuital designs and for future deployment. Unlike prior high-level formulations, this work decomposes the algorithm into the Clifford+T universal gate set, enabling precise resource accounting in terms of logical qubits, gate counts, and circuit depths. We present a comprehensive resource analysis that shows a logarithmic depth scaling with respect to the dataset size, confirming the exponential speed-up obtained in query complexity. We estimate the resources to apply the method to a real-world 2024 bank transaction dataset for fraud detection. Estimates reveal that current hardware limitations—particularly the absence of scalable, low-overhead QRAM devices—dominate the cost, requiring O(10^10) logical qubits for utility-scale deployment. This work provides concrete circuit-level designs for AD establishes concrete and identifies key bottlenecks, identifying concrete hardware performance targets for a quantum advantage in real-world AD tasks.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_10_Guerrini_Tesi_01.pdf
accessibile in internet per tutti
Descrizione: Testo completo della tesi
Dimensione
2.91 MB
Formato
Adobe PDF
|
2.91 MB | Adobe PDF | Visualizza/Apri |
|
2025_10_Guerrini_Executive_Summary_02.pdf
accessibile in internet per tutti
Descrizione: Executive summary della tesi
Dimensione
2.5 MB
Formato
Adobe PDF
|
2.5 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/10589/243980