Unsupervised Machine Learning with Adiabatic Quantum Computing

Abstract

In recent years, quantum computing has emerged as a promising paradigm for solving combinatorial optimization problems that are intractable for classical algorithms. This thesis investigates the application of quantum-inspired methods to unsupervised clustering tasks using adiabatic quantum computing (AQC) principles. The research concentrates on two prominent clustering techniques: K-Means and Minimum Spanning Tree (MST)-based clustering, both implemented in their classical and QUBO-based forms. Simulated annealing is used to mimic quantum annealing behavior due to limited access to real quantum hardware. A series of experiments were conducted on standard benchmark datasets, including Iris, Blobs, and Moons, to compare the performance of classical and QUBO-based approaches using evaluation metrics. The results reveal that while QUBO clustering methods can achieve competitive performance, they are highly sensitive to penalty weight parameters, and their effectiveness diminishes as dataset size increases. These results show important limitations related to scalability and parameter tuning in current QUBO formulations. Even with these challenges, the study illustrates the capability of quantum-inspired clustering and sets the stage for future research. Opportunities for improvement include hybrid quantum-classical approaches, adaptive penalty tuning, and more efficient QUBO encodings. This work helps to the growing of research at the intersection of quantum computing and machine learning, and lays the groundwork for more robust quantum clustering frameworks.