Archer Materials (ASX:AXE) has completed the next stage of its quantum machine learning (QML) fraud detection project by successfully testing and benchmarking its quantum neural network (QNN) model on publicly available financial fraud datasets.
The early models performed on par with the best classical models used in benchmark tests when run on a qubit simulator.
Archer said the model correctly identified 118 fraudulent transactions, with only one false positive.
The company also successfully ran the model on IQM Garnet, a commercial superconducting quantum computer accessible via AWS Braket.
Data set is ready
Archer began the current phase in March after completing dataset preparation and transitioning the project to QML simulations and benchmarks.
The QNN workflow was developed and evaluated using a public financial fraud dataset containing over 280,000 transaction records.
Archer used dimensionality reduction and data balancing techniques to process the dataset, allowing it to operate within the constraints of current quantum computing.
This project used a step-by-step experimental framework covering qubit selection studies, feature map optimization, benchmarking against classical machine learning approaches, and quantum noise analysis.
Minimal false alerts logged
In a simulator test environment, the QNN model was shown to be able to identify fraud with low false alerts. This is one of the biggest practical challenges in fraud detection that can increase costs and degrade customer experience.
The selected model was stable under low levels of simulated quantum noise and showed only a slight performance degradation at moderate noise levels.
Performance degraded significantly at higher noise levels, providing Archer with information about the hardware quality and noise immunity required for future practical QML applications.
The company said the research identified a high-performance quantum architecture, established a reproducible benchmark framework, and provided insights into scaling, deployment constraints, and noise immunity.
IQM hardware verification
Archer also tested the QNN model on IQM Garnet, a 20-qubit superconducting quantum computer available through AWS Braket.
Hardware validation detected 18 out of 19 fraudulent transactions in the test set, demonstrating that the model can work on commercial quantum hardware.
Although the actual hardware experiment had a higher false positive rate than the simulator test, the results provided valuable validation under current quantum computing conditions, Archer said.
Chief executive Dr Simon Ruffel said the simulator and hardware results provided an important validation step.
“These results demonstrated that the QML approach can achieve strong fraud detection performance while operating within the constraints of current quantum computing systems,” said Dr. Raffel.
Target of prototype
Archer’s research collaboration agreement with the Commonwealth Scientific and Industrial Research Organization (CSIRO) remains important.
“The research collaboration agreement with CSIRO forms part of Archer’s strategy to explore the practical application of quantum computing technology and support future commercialization opportunities in data-intensive industries,” Dr Raffel said.
“Fraud detection is a relevant use case for QML, as banks and payment providers need to quickly analyze large amounts of transaction data while reducing both missed fraud and false alarms.”
Archer said further work will include larger datasets, additional classical benchmarks, iterative trials and further hardware validation before evaluating a commercial deployment path.
The company aims to have a complete QML prototype by the end of this year.
