First quantum machine learning applications on an on-site room-temperature quantum computer

Aug 5, 2026

Upload by Hazel

Demonstrates a quantum machine learning algorithm running for the first time on an on-site room-temperature quantum computer. A two-qubit device installed at the Pawsey Supercomputing Centre in Perth solves multi-class classification on unseen 2D data points. The Universal Quantum Classifier model uses a data re-uploading framework, trained on an ideal quantum simulator with Adam, without noise models or device-specific insights. The optimized circuit is deployed via a single XYX decomposition, yielding accuracies comparable to ideal state-vector simulations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages7

Reading Time18 min

AccessFree to download

Introduction

QML categories and Quantum Utility

CQ focus and Universal Quantum Classifier

Experimental setup and deployment

arXiv :2312 . 11673v1 [ quant-ph] 18 Dec 2023 First quantum machine learning applications on an on-site room-temperature quantum computer Nils Herrmann,∗ Daanish Arya, Florian Preis,† and Stefan Prestel Quantum Brilliance GmbH Colorado Tower Industriestr. 4 70565 Stuttgart, Germany Mariam Akhtar, Marcus W. Doherty, Pascal Macha, and Michael L. Walker Quantum Brilliance Pty Ltd 60 Mills Road Acton ACT 2601, Australia (Dated: December 20, 2023) We demonstrate – for the first time – the application of a quantum machine learning (QML) algorithm on an on-site room-temperature quantum computer. A two-qubit quantum computer installed at the Pawsey Supercomputing Centre in Perth, Australia, is used to solve multi-class classification problems on unseen, i.e. untrained, 2D data points. The underlying 1-qubit model is based on the data re-uploading framework of the universal quantum classifier [1] and was trained on an ideal quantum simulator using the Adam optimiser [2] . No noise models or device-specific insights were used in the training process. The optimised model was deployed to the quantum device by means of a single XYX decomposition leading to three parameterised single qubit rotations. The results for different classification problems are compared to the optimal results of an ideal simulator. The room-temperature quantum computer achieves very high classification accuracies, on par with ideal state vector simulations. I. INTRODUCTION Machine learning (ML) is a highly anticipated subfield of artificial intelligence (AI) that aims to find or“learn” optimised models without explicit instructions [7] . The aim is to create models that can effectively predict or classify examples outside the training data in problems where the underlying mathematical correlations are either unknown or too complex for conventional approaches. Here, special care needs to be taken to compile training data sets that ensure a sufficiently general model [8] . Especially in light of the newly developed generative ML models, such as the GPT-4 large language model[9], concerns have been raised addressing their increasing training demands both in energy consumption and carbon footprint [10, 11] . The advent of quantum computing has created high hopes for the field of ML by utilising the exponentially growing n-qubit Hilbert space as the addressable feature space in ML pipelines. Recently, it has been shown that quantum ML (QML) methods are capable to efficiently generalise from fewer training data[12], can achieve better prediction accuracies[13], and possess an inherent advantage in handling quantum data[14] compared to their classical competitors, sparking hopes to arrive at a practical quantum advantage – recently coined as Quantum Utility[15] . Schuld and Petruccione [3] separated QML methods into four distinct categories as shown in subfigure 1a.[10, 11] Here, CC refers to approaches where classical ∗ [nils.herrmann@quantum-brilliance.com](nils.herrmann@quantum-brilliance.com)[ ](nils.herrmann@quantum-brilliance.com)† [f.preis@quantum-brilliance.com](f.preis@quantum-brilliance.com) data is processed using classical algorithms. This includes standard classical ML approaches but in the context of QML refers to those using only quantum-inspired classical algorithms such as tensor network methods. These have recently been used to train ML methods affected by the big data problem in high-energy physics [16] . The QC category involves QML methods that learn from quantum data and apply classical algorithms. Prominent examples include ML-based quantum state tomography [17] and classically distinguishing between quantum states [18, 19] . Of most interest, in the pursuit of Quantum Utility [15], are the CQ and QQ methods, which learn from either conventional or quantum data and train a quantum model, typically a parameterised quantum circuit. Tremendous advances in this field have been reported in the literature over the last year

FAQs

What is demonstrated in the paper?

The paper demonstrates the application of a quantum machine learning algorithm on an on-site room-temperature quantum computer, executing multi-class classification on unseen 2D data points.

How was the underlying model trained before deployment?

The Universal Quantum Classifier model was trained on an ideal quantum simulator using the Adam optimizer, based on the data re-uploading framework, without using noise models or device-specific training insights.

How was the trained model implemented on the room-temperature quantum device?

The optimized model was deployed to the quantum device using a single XYX decomposition that results in three parameterized single-qubit rotations.

Aug 5, 2026

Upload by Hazel