Quantum Gated Recurrent GAN with Gaussian Uncertainty for Network Anomaly Detection
Wajdi Hammami, Soumaya Cherkaoui, Jean-Frederic Laprade, Ola Ahmad, Shengrui Wang · 2025
Read Quantum Gated Recurrent GAN with Gaussian Uncertainty for Network Anomaly Detection by Wajdi Hammami, Soumaya Cherkaoui, Jean-Frederic Laprade, Ola Ahmad, Shengrui Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-26487.txt.
Introduction
Wajdi Hammami, Soumaya Cherkaoui, Jean-Frederic Laprade, Ola Ahmad, and Shengrui Wang design a quantum tool to spot odd behaviour in network traffic. Finding anomalies in time series matters for security, but quantum machine learning tools are held back by small qubit counts. Their system aims to catch attacks both fast and well.
Hammami and his four co-authors pair a quantum gated recurrent unit with a generative adversarial network, two models that learn by competing. The quantum generator outputs the centre and spread of a Gaussian curve, and a gating step flags suspects by uncertainty, then confirms them with critic scores and rebuild errors. On benchmark data, the method reached a time-aware F1 score near eighty-nine percent, beating classical and quantum rivals. It also kept its form when run on real IBM quantum hardware.
Catalog record
| CREATOR | Wajdi Hammami, Soumaya Cherkaoui, Jean-Frederic Laprade, Ola Ahmad, Shengrui Wang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-26487
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Quantum Gated Recurrent GAN with Gaussian Uncertainty for Network Anomaly Detection Wajdi Hammami, Soumaya Cherkaoui, Jean-Frederic Laprade, Ola Ahmad, Shengrui Wang Abstract Anomaly detection in time-series data is a critical challenge with significant implications for network security. Recent quantum machine learning approaches, such as quantum kernel methods and variational quantum circuits, have shown promise in capturing complex data distributions for anomaly detection but remain constrained by limited qubit counts. We introduce in this work a novel Quantum Gated Recurrent Unit (QGRU)-based Generative Adversarial Network (GAN) employing Successive Data Injection (SuDaI) and a multi-metric gating strategy for robust network anomaly detection. Our model uniquely utilizes a quantum-enhanced generator that outputs parameters (mean and log-variance) of a Gaussian distribution via reparameterization, combined with a Wasserstein critic to stabilize adversarial training. Anomalies are identified through a novel gating mechanism that initially flags potential anomalies based on Gaussian uncertainty estimates and subsequently verifies them using a composite of critic scores and reconstruction errors. Evaluated on benchmark datasets, our method achieves a high time-series aware F1 score (TaF1) of 89.43% demonstrating superior capability in detecting anomalies accurately and promptly as compared to existing classical and quantum models. Furthermore, the trained QGRU-WGAN was deployed on real IBM Quantum hardware, where it retained high anomaly detection performance, confirming its robustness and practical feasibility on current noisy intermediate-scale quantum (NISQ) devices. I Introduction Anomaly detection in time-series data plays a vital role in monitoring the behavior of complex and dynamic communication systems, where temporal dependencies heavily influence traffic patterns [ 1 ] . These anomalies—subtle and often transient deviations from normal network behavior—can indicate serious issues such as cyberattacks or system failures. Unlike static datasets, time-series network data presents unique challenges, since irregularities must not only be flagged, but also explained within the evolving temporal context of network activity. Traditional statistical approaches often fall short in capturing non-linear dependencies or adapting to evolving patterns over time, necessitating more sophisticated learning-based methods to model the temporal and structural intricacies of modern networks. In recent years, machine learning has become a cornerstone in solving network anomaly detection problems across various domains [ 32 ] . By learning patterns directly from data, machine learning models can adapt to complex, high-dimensional environments where manual rule-setting is impractical. Techniques such as autoencoders [ 3 ] , recurrent neural networks [ 26 ] , and generative models [ 16 ] have shown strong capabilities in modeling normal behavior and flagging deviations. These models are particularly valuable in settings where data distributions shift over time or labeled anomalies are scarce. As a result, machine learning based anomaly detection techniques has become integral component for modern network security systems [ 23 ] . Alongside the advancements in classical machine learning, quantum computing has emerged as a promising frontier for solving complex computational problems. Quantum machine learning (QML), a fusion of quantum computing principles with data-driven learning, is attracting increasing attention for its potential to outperform classical models in specific tasks [ 17 ] , thanks to its leveraging quantum phenomena such as superposition and entanglement to encode rich feature spaces and capture complex correlations that are challenging for classical models. While empirical evidence on large-scale real-world datasets remains limited, recent studies suggest that hybrid quantum-classical models may offer advantages in scenarios involving high-dimensional data, limited samples, or probabilistic modeling [ 11 ] —challenges commonly encountered in network anomaly detection tasks. This growing synergy between quantum computing and machine learning opens new avenues for addressing the complexity of network anomaly detection in ways that classical methods may struggle to achieve. In this work, we focus on the problem of anomaly detection within network time-series data—a setting where temporal structure plays a crucial role in identifying irregularities. Given the sequential nature of the data as well as the need to capture contextual patterns and handle uncertainty, we propose to leverage the strengths of both classical machine learning and quantum computing. Specifically, we explore the integration of recurrent neural architectures, known for their temporal modeling capabilities, with quantum-enhanced learning mechanisms that offer expressive generative potential [ 20 ] . This convergence allows us to build a hybrid model tailored for detecting anomalies in network time-series, where subtle and context-dependent deviations must be identified with high precision. We address several key limitations that arise in time-series network anomaly detection. While traditional GAN architectures can model data distributions, they often struggle to capture complex temporal dependencies inherent in sequential data. To address this limitation, we integrate Gated Recurrent Units (GRUs), which are well-suited for modeling sequential patterns due to their efficient parameterization. To enrich the model’s expressive capacity, we introduce Variational Quantum Circuits (VQCs) that embed the latent representations into richer Hilbert spaces, enabling the capture of complex feature interactions [ 4 ] . Since directly encoding large input dimensions into quantum circuits remains constrained by qubit availability, we employ Successive Data Injection [ 21 ] (SuDaI), which allows multiple segments of temporal data to be sequentially injected into the quantum circuit, thereby efficiently utilizing limited quantum resources. Finally, we combine these components within a hybrid GAN architecture, and introduce a multi-metric anomaly detection pipeline that leverages predictive uncertainty, adversarial feedback, and reconstruction-based criteria to robustly detect anomalies with greater contextual sensitivity. We have evaluated our method using the HAI (HIL-based Augmented Industrial Control Systems) dataset [ 29 ] , which provides multivariate time-series data collected from a network-integrated industrial control system under both normal operations and staged network-based cyber-attacks. Performance is assessed using the extended time-series aware F1 score (TaF1) metric [ 19 ] , which accounts for anomaly detection accuracy over temporal intervals. Our approach achieves an TaF1 score of 88.63%, substantially surpassing existing methods. Furthermore, the trained model was successfully executed on real IBM Quantum hardware without retraining, demonstrating that the proposed quantum architecture is not only effective in simulation but also operationally viable on current NISQ devices. The paper proceeds as follows: Section II covers related literature in classical and quantum anomaly detection methods. Section III outlines our proposed model’s architecture. Section IV details the experimental design and results. Section V concludes the paper. II Related works Time-series anomaly detection (TSAD) has gained significant attention across domains such as cybersecurity, industrial monitoring, and IoT systems [ 5 ] , where early detection of irregular behavior is crucial. Traditional statistical and clustering methods [ 6 ] offer basic capabilities but struggle with high-dimensional, non-linear, and noisy data. Deep learning has advanced the field through generative and prediction-based models, particularly by using RNNs and GANs [ 25 , 10 , 31 , 32 ] . These models have shown success in capturing complex temporal patterns but often face challenges related to data imbalance, threshold setting, and explainability [ 28 , 33 ] . GAN-based models [ 34 ] represent a prominent approach in unsupervised anomaly detection for network systems. These models learn the distribution of normal data and detect anomaly by measuring deviations. Despite recent progresses, stabilizing GAN training remains challenging due to issues such as mode collapse and unstable convergence. Methods such as the Wasserstein loss [ 14 ] have been proposed to address these challenges, allowing to improve the training stability without fully resolving the problem. Simultaneously,