Chiatto, Angela (2025) Computational Intelligence for Variational Quantum Algorithms. [Tesi di dottorato]

Documento PDF PhD_thesis_Angela_Chiatto_final_version.pdf Visibile a [TBR] Amministratori dell'archivio Download (5MB) | Richiedi una copia

Abstract

The present work is focused on the design and development of Computational Intelligence (CI) techniques aimed at enhancing the performance of Variational Quantum Algorithms (VQAs). VQAs, which are trainable quantum algorithms, are key candidates for achieving quantum advantage in areas such as optimization and machine learning. However, their performance is hindered by issues of trainability and efficiency. The benefits of integrating VQAs and CI methods have been demonstrated on both benchmark and real-world problems. A major issue with VQAs is their trainability, particularly due to the barren plateau phenomenon, where large-scale problem spaces become flat with narrow gorges and featureless, making conventional gradient-based optimization methods ineffective. This issue is compounded by the limited quantum resources available and the noise present in current quantum devices, known as Noisy Intermediate-Scale Quantum (NISQ) devices. Evolutionary Algorithms (EAs) have been proposed to overcome these limitations in both quantum classification and quantum combinatorial optimization problems, as they are robust and flexible heuristic search methods that do not rely on gradient information, are robust to the noise, and can escape local minima. Additionally, EA-based VQAs have been applied to solve real-world problems, particularly in the context of power networks and healthcare. In addressing the efficiency of quantum resources, fuzzy clustering has been proposed to reduce the complexity of the Quantum Approximate Optimization Algorithm (QAOA), a key quantum algorithm for combinatorial optimization. A combination of fuzzy clustering with Genetic Algorithms has also been proposed to enhance QAOA training, further improving its performance. Based on these advancements, two software tools have been developed to facilitate the integration of CI techniques into VQA framework. The first, named EVOVAQ, is a Python library designed to simplify VQA training with EAs. The second tool, named QCweb, is a web-based application with both educational and experiment-making functions, enabling users to build and train quantum classifiers using EAs.

Downloads

Actions (login required)

Modifica documento