Related Experiment Video
Updated: Aug 11, 2026
11:45
Experimental Methods for Trapping Ions Using Microfabricated Surface Ion Traps
Published on: August 17, 2017
Benchmarking a Tunable Quantum Neural Network on Trapped-Ion and Superconducting Hardware
Djamil Lakhdar-Hamina1, Xingxin Liu1, Richard Barney1
1University of Maryland, College Park, Joint Quantum Institute and Department of Physics, Maryland 20742, USA.
Physical Review Letters
|August 10, 2026
Summary
We demonstrate a quantum neural network for image classification on real quantum hardware. Introducing measurement uncertainty via an interpolation parameter enhances classification performance on noisy intermediate-scale quantum devices.
Area of Science:
- Quantum Computing
- Machine Learning
- Artificial Intelligence
Background:
- Quantum neural networks (QNNs) offer a novel approach to machine learning by leveraging quantum phenomena.
- Implementing QNNs on current quantum hardware presents challenges due to noise and limited qubit connectivity.
Purpose of the Study:
- To implement and evaluate a QNN for image classification on trapped-ion and superconducting quantum computers.
- To investigate the effect of classical-quantum regime interpolation on classification performance.
- To analyze the impact of physical noise on QNN inference and compare it with idealized simulations.
Main Methods:
- A QNN architecture was designed for Modified National Institute of Standards (MNIST) image classification.
- Feedforward was implemented using qubit rotations conditioned on measurement outcomes.
- The network was trained classically, with inference performed on experimental quantum hardware (trapped-ion and IBM superconducting).
- A tunable interpolation parameter was used to bridge classical and quantum computational regimes.
Main Results:
- Moderate interpolation values improved classification performance by introducing beneficial measurement uncertainty.
- Deviations from idealized simulations were observed for borderline images, attributed to physical noise and landscape fluctuations.
- Noise benchmarking was performed by inserting additional gate pairs into quantum circuits.
- The experimental QNN demonstrated potential for image classification tasks on noisy intermediate-scale quantum devices.
Conclusions:
- The study successfully implemented and tested a QNN for image classification on distinct quantum computing platforms.
- Measurement uncertainty, controlled by interpolation, can enhance QNN performance in noisy environments.
- Physical noise significantly impacts QNN inference, highlighting the need for error mitigation strategies.
- The findings support the development of more complex QNNs and explore pathways toward classically intractable quantum machine learning architectures.
