Implementation and empirical evaluation of a quantum machine learning pipeline for local classification.

Area of Science:

• Quantum Computing
• Machine Learning
• Artificial Intelligence

Background:

• Quantum resources are limited, hindering quantum machine learning (QML) model development.
• Quantum locality techniques, like k-nearest neighbors (k-NN), can optimize QML by focusing on relevant data neighborhoods.
• Quantum k-NN has not been previously integrated as a preliminary step in other QML models, unlike its classical counterpart which shows performance enhancements.

Purpose of the Study:

• To propose and evaluate the use of quantum locality techniques to reduce QML model size and improve performance.
• To implement and empirically assess a QML pipeline for local classification.
• To investigate the effectiveness of quantum locality in the QML domain.

Main Methods:

• Development of a QML pipeline using Qiskit, incorporating a quantum k-NN algorithm and a quantum binary classifier.
• Python implementation of the QML pipeline for local classification.
• Extensive empirical evaluation of the proposed quantum pipeline.

Main Results:

• The quantum pipeline demonstrated accuracy equivalence to its classical counterpart under ideal conditions.
• The study validated the applicability of locality techniques within the quantum machine learning realm.
• The specific quantum k-NN implementation exhibited high sensitivity to probability fluctuations, and classical methods like random forest showed superior performance.

Conclusions:

• Quantum locality techniques are viable for QML, offering potential for model optimization.
• The sensitivity of quantum k-NN to noise necessitates further research and development.
• Classical machine learning methods remain competitive and may offer better performance in certain scenarios.