Modeling Quantum Autoencoder Trainable Kernel for IoT Anomaly Detection

TL;DR

The paper tackles real-time anomaly detection in high-dimensional IoT traffic under limited compute budgets, where classical methods struggle at the edge. It proposes a hybrid quantum framework combining a coherence-driven quantum autoencoder with a trainable quantum kernel and QSVC, using amplitude encoding and SWAP-test fidelity to learn discriminative latent representations. The model is trained with a COBYLA optimizer, and kernel parameters are optimized via SPSA, with depolarizing noise serving as implicit regularization. Empirical results on Bot_IoT, IoT23, and KDD99 show high accuracies in simulation and robust performance on IBM hardware, surpassing prior QAE-based baselines, especially under noise. The work demonstrates practical quantum advantage for cyber threat detection and suggests hardware-ready quantum ML pipelines for constrained IoT environments.

Abstract

Escalating cyber threats and the high-dimensional complexity of IoT traffic have outpaced classical anomaly detection methods. While deep learning offers improvements, computational bottlenecks limit real-time deployment at scale. We present a quantum autoencoder (QAE) framework that compresses network traffic into discriminative latent representations and employs quantum support vector classification (QSVC) for intrusion detection. Evaluated on three datasets, our approach achieves improved accuracy on ideal simulators and on the IBM Quantum hardware demonstrating practical quantum advantage on current NISQ devices. Crucially, moderate depolarizing noise acts as implicit regularization, stabilizing training and enhancing generalization. This work establishes quantum machine learning as a viable, hardware-ready solution for real-world cybersecurity challenges.

Figures (5)

• Figure 1: Overall pipeline of the proposed quantum autoencoder–based anomaly detection framework. (i) data preprocessing, including standardization, normalization, and power-of-two padding (ii) quantum autoencoder training, where RawFeatureVector amplitude encoding, a trainable feature map, and the autoencoder circuit are optimized via a sampler QNN and cost function to learn compressed latent representations and a trainable quantum kernel; and (iii) evaluation, in which a QSVC classifier uses the learned quantum kernel to perform anomaly detection.
• Figure 2: Training loss curves for the quantum autoencoder on Bot_IoT, IoT23, and KDD datasets under noiseless and depolarizing noise conditions showing convergence of model acreoss iteraions.
• Figure 3: Normalized gradient for autoencoder and kernel training phases under depolarizing noise as well for Bot_IoT, IoT23, and KDD
• Figure 4: Coefficient of variation of loss estimates for the QSVC autoencoder and kernel phases, comparing noisy and noiseless configurations.(a) and (b) show the variability during the autoencoder phase, while (c) and (d) depict the kernel phase across training iterations.
• Figure 5: Performance of the QAE–QSVC model on IBM Quantum hardware (ibm_fez) for Bot_IoT and IoT23, demonstrating the practical feasibility of deploying quantum autoencoder-based anomaly detection on current NISQ devices.