Researchers Estimate Fidelity and Save Computation with New Quantum Scheduling System

Assigning quantum circuits to specific processors in a multiprocessor system presents key challenges due to varying qubit quality and connectivity. A framework developed at Technical University of Munich predicts how well each circuit will perform on different quantum processing units before compilation begins. It selects the most appropriate processor for each task, approximating optimal assignment while reducing computational demands compared with testing every possibility.

The team has created a new system to manage quantum computing tasks across multiple processors; these Quantum Processing Units, or QPUs, vary in quality and capability. The framework accurately predicts how well each calculation will perform on different devices *before* compiling the program which reduces wasted effort. By estimating performance beforehand using a type of artificial intelligence called a Graph Neural Network, circuits can then be assigned to the most suitable processor for reliable results while balancing computational demands.

Efficiently distributing tasks across multiple processors is vital in scaling quantum computing. Modern High Performance Computing-Quantum Computing platforms now offer several Quantum Processing Units, essentially the quantum equivalent of a computer processor where calculations use qubits instead of bits, each with varying capabilities. This prediction relies on a Graph Neural Network, a type of artificial intelligence that learns relationships between data points much like analysing connections within social networks and can accurately forecasting performance truly overcome the complexities inherent in noisy intermediate-scale quantum devices and unlock efficient multiprocessor operation.

Graph neural networks predict quantum circuit fidelity for optimised multiprocessor allocation

Scientists have developed a new scheduling framework capable of approximating exhaustive fidelity-based quantum circuit assignment while reducing computational resource demands by up to an order of magnitude compared to brute-force methods. This breakthrough overcomes a key threshold previously hindering efficient multiprocessor operation: accurately predicting circuit performance *before* compilation across diverse Quantum Processing Units or QPUs, the quantum equivalent of computer processors.

A Graph Neural Network forms the core of this system; it learns relationships within data to forecast expected execution fidelity on each device and enables intelligent task distribution with tunable prioritisation between accuracy and parallel processing capabilities.

Further validation took place at Technical University of Munich through emulation using data from real IQM superconducting devices, revealing an ability to predict post-compilation circuit fidelities with low error rates closely matching assignment performance achieved via exhaustive testing. Specifically, the framework demonstrated capacity for tunable prioritisation between accuracy and parallel processing, allowing users to adjust how strongly fidelity influences device selection.

Tunable prioritisation offers flexibility for diverse computing needs by balancing high accuracy against maximising parallel processing potential. The Graph Neural Network accurately assessed expected outcomes on each Quantum Processing Unit or QPU, enabling intelligent distribution across varied hardware configurations including differences in qubit count, connectivity and gate characteristics.

Graph neural networks predict quantum circuit fidelity during multiprocessor task distribution

The scheduling framework provides a promising solution to maximise fidelity when distributing quantum circuits across multiple processors; however, approximating an optimal assignment alone is insufficient. Previous work attempted similar pre-compilation predictions with tools like QuEst and MQT Predictor, but this approach introduces a novel method utilising Graph Neural Networks, a machine learning type particularly suited for analysing connections between components, for fidelity estimation. Although accuracy or computational efficiency compared with these alternatives wasn’t detailed, the delivery of another perspective remains valuable.

Intelligent task distribution becomes possible as the predictive analysis allows quantum computers to assign tasks across multiple processors, prioritising those best equipped to handle them while minimising errors. Avoiding wasted computational effort inherent in testing every device assignment also enables exploration of trade-offs between speed and reliability; it represents significant progress towards efficient quantum computation. By predicting circuit performance before compilation, the process translating instructions into machine code, employing a Graph Neural Network sharply reduces computational demands beyond simply saving resources.

The research demonstrated a new framework that predicts how well quantum circuits will perform on different Quantum Processing Units or QPUs prior to running calculations. This prediction uses a Graph Neural Network to assess fidelity, essentially, accuracy, across varied hardware with differing qubit counts and connectivity. The system then schedules tasks intelligently, balancing execution fidelity against parallel processing potential, approximating an optimal assignment without compiling each circuit on every device. Consequently, this approach offers increased efficiency by reducing wasted computation when distributing workloads among multiple processors.

More like this

Quantum Research NewsUtrecht University Cuts Simulation Times with New Symmetry Tool

Quantum Research NewsGame Shows Limits to Testing Entangled Qubits with Six Dimensions

Quantum Research NewsStony Brook Team Bounds Noisy Entanglement Thresholds

Quantum Research NewsResearchers Achieve Chemical Accuracy Using Quantum Error Mitigation