A Machine Learning Driven Approach to Quantum Network Tomography.

Quantum Networks are interconnected quantum systems that exchange quantum information. Creating such networks is a foundational step towards realizing a quantum internet, which is essential for meeting the demands of large-scale quantum computing. The biggest issue quantum networks face is quantum information’s inherent sensitivity to noise. Characterizing how this noise affects networks is necessary to establish reliable transmission. Quantum Network Tomography (QNT) aims to address this problem by analyzing how quantum states evolve across the network. The Quantum Fisher Information Matrix (QFIM) is a measure of how well a QNT protocol captures network related information. Designing protocols that maximize the QFIM remains a challenging and unsolved problem, as traditional approaches are often computationally demanding and struggle with the complexity of realistic quantum networks. Machine learning presents a promising alternative.

Here, we present how we can utilize  machine learning approaches for quantum network tomography. We created multiple reinforcement learning agents (RLA’s) to design optimal QNT protocols that maximize the QFIM. In particular, we show how these RLA’s adapt to different network topologies, error models, and constraints, and how the efficiency and accuracy of this machine learning driven approach compares to other manual and computational methods.