article · Applied Artificial Intelligence
Quantum machine learning holds potential for cybersecurity, but running models on current noisy intermediate-scale quantum hardware remains challenging due to limited resources. To address this, research evaluated three data-encoding feature maps: IQP, Angle, and Fourier embeddings, using quantum neural networks across cybersecurity datasets. The most effective encoding was integrated into data re-uploading and hybrid quantum-classical models. This led to the creation of AdaReQ, a model that dynamically alternates between two embeddings to fit dataset characteristics. Evaluated on three benchmark datasets, AdaReQ achieved 96.1 percent accuracy on BODMAS, surpassing state-of-the-art quantum benchmarks, and 98.4 percent on HIKARI-2021, outperforming recent classical models, while reaching 86.2 percent on EMBER. Remarkably, AdaReQ requires only four qubits and 24 trainable parameters, significantly lowering the hardware requirements compared to alternatives.
Current quantum computers are severely constrained by physical noise and limited qubit numbers, making complex threat analysis difficult to deploy. By demonstrating that high detection accuracy can be achieved using very small quantum circuits, this development shows that practical cyber defence tools can run efficiently on present-day quantum hardware without waiting for large-scale, fault-tolerant machines.
The method could enable resource-efficient threat detection for network security providers and security operations centres looking to integrate quantum algorithms. It remains at an early, laboratory-tested stage, having been evaluated strictly on benchmark datasets such as EMBER, BODMAS, and HIKARI-2021, and has not yet been demonstrated in live, operational production environments.
AI-generated from the published abstract. Always read the original work before citing.
Quantum machine learning (QML) presents promising opportunities for advancing cybersecurity, yet its implementation on Noisy Intermediate-Scale Quantum (NISQ) hardware remains challenging. This study first evaluates three feature maps (IQP, Angle, and Fourier embeddings) using Quantum Neural Networks (QNNs) to identify the most effective encoding strategy for each cybersecurity threat dataset. The selected feature map is then applied in two additional QML models: a data re-uploading approach and a hybrid quantum–classical model based on observable construction. We introduce AdaReQ, a novel architecture that dynamically alternates between two embeddings to adapt to dataset-specific characteristics. Evaluated on the EMBER, BODMAS, and HIKARI-2021 datasets, AdaReQ achieves 86.2% accuracy on EMBER (vs. 92.5% state-of-the-art QML), surpasses the state-of-the-art QML benchmark on BODMAS with 96.1% accuracy, and outperforms recent classical baselines on HIKARI-2021 with 98.4% accuracy, all while using only 4 qubits and 24 trainable parameters, compared to alternative models requiring more than twice the parameters and four times the qubits.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1080/08839514.2026.2721305
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
