NeTS: Small: Surrogate-Assisted Full-Stack Optimization of Quantum Networks with Reinforcement Learning

About This Grant

Quantum networks promise to revolutionize technology by enabling provably secure communication, ultra-precise sensors for scientific discovery, and new forms of distributed and blind quantum computing. However, current quantum hardware is noisy, error-prone, and inefficient, making the reliable distribution of entanglement -- the key resource powering these applications -- a formidable challenge. This project seeks to overcome these hurdles by developing a novel, comprehensive framework to design and operate quantum networks with maximum efficiency. By creating intelligent control policies that are co-designed with the underlying hardware, this work will enable near-term quantum systems to perform tasks that are currently out of reach. This research serves the national interest by accelerating the development of a secure quantum communication infrastructure, a cornerstone for national defense and economic prosperity. The project’s advancements will also support scientific progress by enabling new instruments, such as quantum-enhanced telescopes, and will foster economic welfare through applications in drug discovery and materials science. The project will produce open-source software tools, benefiting the entire research community. Furthermore, it includes significant educational components, developing new university curricula, supporting undergraduate research, and creating pathways for students to enter the quantum information science workforce. This project's central goal is to develop a full-stack framework for the design and optimization of control policies for distributed quantum systems. The approach integrates analytical modeling, high-fidelity simulation, and machine learning to discover robust, hardware-aware protocols. The methodology begins with analytically inspired policy ansatzes that are then refined using Reinforcement Learning (RL) within a realistic simulation environment that captures complex hardware characteristics. To overcome the high computational cost of training, the framework will leverage a combination of high-fidelity simulations for validation and newly developed, efficient surrogate models for both quantum state dynamics and network-level control logic. Key technical innovations include moving beyond simplistic noise models to more faithful representations of biased Pauli noise, amplitude loss, and coherent errors. The project will develop application-specific utility functions for key use cases, including quantum key distribution, quantum-enhanced interferometry, and blind quantum computing, enabling optimization for true performance rather than proxy metrics. The research will explicitly address critical challenges in applying RL to quantum networks, such as state-space explosion, sparse and non-additive rewards, and the generation of deployable policies that operate with local knowledge. The project's outcomes will be a set of optimized entanglement distribution protocols and a principled, scalable methodology for evaluating quantum network utility. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.