Quantum neural network for entanglement purification and noise mitigation

Abstract

Quantum entanglement serves as a foundational resource for quantum communication, distributed computing, and secure information processing. However, real-world quantum systems are highly susceptible to noise and decoherence, which degrade entanglement fidelity and hinder practical scalability in near-term noisy intermediate-scale quantum (NISQ) devices and quantum networks. Traditional entanglement purification protocols, such as BBPSSWand DEJMPS, often rely on fixed, hand-crafted operations that struggle with non-identical input states, heterogeneous noise models, and high resource overhead. This thesis proposes a novel hybrid Quantum Neural Network (QNN) framework that integrates parameterized quantum circuits with classical neural architectures to learn adaptive entanglement purification and noise mitigation strategies directly from noisy quantum data. By leveraging quantum superposition, entanglement, and variational optimization, the QNN models map ensembles of degraded entangled states to higherfidelity outputs while minimizing qubit consumption and adapting in real time to realistic noise channels (e.g., amplitude damping, phase damping, and memory decoherence). Simulations demonstrate that this approach achieves superior fidelity gains, higher success probabilities, and improved throughput compared to conventional methods, o!ering a scalable pathway toward robust quantum networks and modular quantum computing. This work advances the intersection of quantum machine learning and entanglement manipulation, providing new insights into data-driven protocol design for noisy quantum environments.