Back to search

Democritus University of Thrace (DUTH)

Ανάπτυξη μοντέλων μηχανικής μάθησης κβαντικής υπολογιστικής και εφαρμογές

Abstract

dc:description

Computing performance has historically advanced through continued transistor scaling and increased integration density. According to Gordon Moore’s observation in 1965, the number of transistors per integrated circuit doubles every two years. This trend was confirmed for decades and contributed to the growth of computational capabilities that helped artificial intelligence and large-scale data processing grow. However, the ongoing miniaturization of transistors, which today have reached 2 nm, creates serious challenges. At dimensions on this scale and below, the behavior of semiconductors can no longer be described by classical physics but by the laws of quantum mechanics, and thus different computing approaches are necessary. In this direction, quantum computers aim to exploit capabilities described by quantum mechanics to perform computational procedures. From Richard Feynman’s theoretical foundations in the 1980s to modern commercial quantum computers, this field has seen rapid development. Indicatively, Google’s Willow quantum processor completed a benchmark problem in less than five minutes, for which a supercomputer would need more than 1025 years. This progress has triggered the fast growth of quantum computing and the creation of additional research fields, such as quantum machine learning and quantum deep learning, with the goal of leveraging quantum advantages to solve complex problems. Despite this momentum, the field faces major challenges. Noise in quantum computations causes instability in process outcomes, making reproducibility difficult, while available quantum resources remain limited, with the result that most existing research is carried out on simulators. These challenges have created gaps both in the diversity of classical algorithms transferred to the quantum domain and in the achievement of quantum model superiority. The present research aims to address these gaps. This thesis presents conducted research with the aim of strengthening the scientific field of quantum machine learning through the investigation and advancement of existing models, focusing on addressing their weaknesses. At the core of the research is the Fuzzy Cognitive Map (FCM). Although this model has evolved in classical machine learning, with multiple architectural variants and applications in pattern recognition, it faces fundamental limitations. The linear nature of its inference mechanism makes it difficult to capture nonlinear relationships in complex data, while its scalability and its application to other data structures, such as images, remain open issues. In quantum machine learning, existing approaches are mainly limited to hybrid forms of optimization, where quantum inspired, algorithms are used to train FCM, without modifying the architecture itself. This research addresses that gap by proposing three models that structurally modify the FCM by incorporating quantum mechanisms. The first model named Quantum Fuzzy Cognitive Map (QFCM) is the quantum version of the classical one, where, in a hybrid form, the classical cognitive map is replaced by a parameterized Variational Quantum Circuit (VQC) that exploits superposition and entanglement for processing and representing causal relationships in data. The second model named Quantum Kernelized FCM (QKFCM), integrates a variational quantum kernel into the inference rule of the classical FCM, inspired by radial basis function networks, replacing the linear influence of each input with a nonlinear similarity measure and giving the model the ability to capture complex relationships among features. The third model implements a deep architecture of the QFCM, combining the quantum FCM model with quantum convolutional neural networks (QFCM-QCNN), enabling the model to process higher dimensional data structures such as images.The research methodology combines a literature review, the design and implementation of quantum models, and extensive experimental evaluation. The experiments examine the performance of the quantum models across various machine learning datasets, as well as their robustness against malicious data. In each case, the corresponding classical architectures were also implemented to allow direct comparison and to document the points where the quantum models are superior. The results confirm that the proposed approaches achieve high performance on certain problems while being competitive with classical methods on others. This dissertation contributes to quantum machine learning by introducing a QFCM model and two FCM-based quantum variants that extend the pattern recognition capabilities of FCM through quantum computational mechanisms. Finally, this research and its methodology, which unifies literature review, architectural study, and experimental evaluation, can serve as a template for future work aiming to transfer classical models to the quantum ecosystem, thereby providing extensive knowledge and information to the research community.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kalampokas, Theofanis
  • Καλαμπόκας, Θεοφάνης

Subjects

dc:subject × 20

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62468

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kalampokas, Theofanis; Καλαμπόκας, Θεοφάνης. Ανάπτυξη μοντέλων μηχανικής μάθησης κβαντικής υπολογιστικής και εφαρμογές. Democritus University of Thrace (DUTH), 2026. http://hdl.handle.net/10442/hedi/62468