Machine Learning Assisted Noise Classification and Mitigation in Quantum Key Distribution Protocol

Abstract

This thesis presents machine learning-based techniques for binary classification and mitigation of noise in Quantum Key Distribution (QKD) protocols. The classification task is carried out in two distinct settings: quantum channels, using simulated density matrix models, and on gate-based quantum computers, using IBM Qiskit implementations of the BB84 and BBM92 protocols. To distinguish between bit-flip, amplitude damping, and depolarizing noise, we employ supervised learning modelS - K-Nearest Neighbors, Gaussian Naive Bayes, and Support Vector Machine which achieve high classification accuracy across both environments. Building on this, we develop a Singular Value Decomposition (SVD) based mitigation strategy, which significantly reduces error in QKD channels. Together, these methods provide a practical and effective framework for enhancing the performance and reliability of quantum communication systems.