Adaptive Optimization of Latency and Throughput with Fidelity Constraints in Quantum Networks Using Deep Neural Networks

Abstract: Quantum networks rely heavily on the establishment of high-fidelity entanglement links, yet achieving target fidelity typically introduces trade-offs between latency and throughput. In this paper, we propose a semi-supervised adaptive purification approach employing Deep Neural Networks (DNNs) to optimize the balance between latency, throughput, entanglement resource utilization and fidelity thresholds in quantum repeater networks. By intelligently predicting the necessary rounds of purification per link, our method dynamically adapts to varying fidelity requirements across different quantum communication use-cases. Through simulations integrating quantum purification, entanglement establishment, and network-level request scheduling, we compare our approach to fixed-round purification and FIFO schemes. We show that the proposed scheme achieves greater flexibility in adjusting final entanglement fidelity levels while minimizing latency and enhances the efficient utilization of entangled Bell pairs. Our results underline the potential of deep learning techniques for achieving adaptive, optimized performance in future quantum networking applications.