Fiddle: Reinforcement Learning Maximizes Quantum Fidelity in Routing, Addressing Noise in Computing Systems

Quantum computing promises transformative advances in fields like optimisation and machine learning, but current quantum devices suffer from errors introduced by environmental noise, limiting their practical application. Hoang M. Ngo, Tamer Kahveci, and My T. Thai from the University of Florida tackle this fundamental challenge by presenting a new framework, FIDDLE, which directly enhances the reliability of quantum circuits during a crucial step called transpilation. This research introduces a novel approach to optimising the routing of quantum information, using a combination of predictive modelling and reinforcement learning to maximise process fidelity, a direct measure of circuit reliability. Unlike existing methods that focus on indirect metrics like circuit complexity, FIDDLE demonstrably improves the accuracy of quantum computations across a range of realistic noise conditions, representing a significant step towards building fault-tolerant quantum computers.

The study addresses the significant challenge of noise in current quantum devices, which limits the potential of quantum computing. Scientists focused on maximizing process fidelity, a comprehensive measure of how closely a quantum state matches its ideal form, offering a more robust assessment of reliability than previous methods. Researchers developed a two-module system to achieve improved fidelity. The first module employs a Gaussian Process-based surrogate model, which accurately estimates process fidelity using limited training data, overcoming the computational expense of exact calculations.

This surrogate model then integrates with a reinforcement learning module, creating an end-to-end framework capable of optimizing routing decisions within quantum circuits. The reinforcement learning component learns to select routing paths that maximize predicted process fidelity, guided by the estimations provided by the Gaussian Process model. This allows the system to efficiently explore possible configurations and identify those yielding the most reliable circuits. Extensive evaluations demonstrate that the surrogate model provides a more accurate estimation of process fidelity compared to other learning-based techniques. Furthermore, the complete FIDDLE framework significantly improves the process fidelity of quantum circuits across a range of noise models. This work establishes a strong foundation in both quantum computing and machine learning, and represents a substantial advancement in the development of robust and reliable quantum computations.

Gaussian Processes Boost Quantum Computation Fidelity

Scientists have developed a new framework, FIDDLE, to significantly enhance the reliability of quantum computations. The core challenge addressed is noise inherent in current quantum devices, which limits the scalability of quantum computing for practical applications. Researchers focused on improving process fidelity, a metric that evaluates the overlap between ideal and actual quantum states, offering a more comprehensive measure of reliability than previous methods. Existing techniques for calculating process fidelity are computationally expensive. To overcome this, the team introduced a Gaussian Process-based surrogate model that accurately estimates process fidelity with a limited number of training samples, addressing the impracticality of large-scale training in existing learning-based techniques.

This surrogate model provides a better estimation of process fidelity, enabling efficient evaluation of circuit reliability. Furthermore, FIDDLE incorporates a reinforcement learning module to optimize the routing stage of quantum circuit transpilation, directly maximizing process fidelity rather than relying on indirect metrics like circuit depth or gate count. This approach represents a breakthrough, as most existing methods focus on optimizing factors indirectly related to reliability. Experiments demonstrate that the end-to-end framework significantly improves the process fidelity of quantum circuits across various noise models, paving the way for more robust and scalable quantum computations.

Fidelity Improvement via Learning and Optimisation

This work presents FIDDLE, a novel learning-based framework designed to enhance the reliability of quantum circuits, a critical challenge in the current era of noisy intermediate-scale quantum (NISQ) computing. Researchers developed a framework that efficiently estimates process fidelity using a Gaussian Process-based surrogate model, even when limited training data is available, and then optimizes gate sequences for improved fidelity through a reinforcement learning module. Rigorous evaluations demonstrate that FIDDLE significantly improves both the accuracy of fidelity estimation and the optimization of circuit fidelity across a range of different noise models. These improvements represent a substantial advance over existing state-of-the-art methods for fidelity estimation and routing optimization. As a result, this research lays a foundation for developing practical and scalable gate-based quantum computing applications.