Comparing Classical and Quantum Deep Learning Techniques for Anomaly Detection of Short-Duration Gamma-Ray Signals
A. Rizzo, N. Parmiggiani, Andrea A. Bulgarelli, Antonio Macaluso, C. Burigana, Eric Burns, Luca Cappelli, V. F. Cardone, Farida Farsian, Savitri Gallego, Giuseppe Murante, Giuseppe Sarracino, R. Scaramella, Francesco Schillirò, V. Testa, T. Trombetti, Dieter H. Hartmann, Carolyn Kierans, Eliza Neights, John A. Tomsick, Andreas C. Zoglauer
Gamma-Ray Bursts (GRBs) are the most luminous explosions in the universe, and are of great interest in astrophysics research. Despite the development of onboard trigger algorithms, significant obstacles remain in identifying weak events and effectively filtering out false detections. These challenges motivate the exploration of innovative technologies such as deep learning and quantum computing to help traditional triggers as a second verification level and increase reliability across the detection pipeline. We present a comparison of a classical and a quantum autoencoder running on simulated quantum hardware, tackling an anomaly detection task in a controlled environment. The goal is to compare their performance, with a focus on resource-constrained scenarios involving limited data and trainable parameters. Both approaches were tested using a simulated set of light curves from the anticoincidence shield of the Compton Spectrometer and Imager (COSI) soft gamma-ray telescope. Short-duration GRB signals were simulated using MEGAlib and then modified through data augmentation techniques. The results indicate that while classical autoencoders reconstruct better, achieving a Mean Squared Error (MSE) as low as 2 . 97 × 1 0 − 5 , the efficiency of the quantum autoencoders is notable with lower numbers of parameters and smaller datasets. Using only 10 trainable parameters and 100 input samples, the quantum model obtained an MSE of 2 . 24 × 1 0 − 4 , outperforming its classical counterpart (MSE of 3 . 81 × 1 0 − 2 with 705 parameters). These results show a promising role for quantum machine learning in astrophysical contexts where data related to GRBs are limited or lightweight model architectures are essential.
Gamma-Ray Bursts (GRBs) are the most luminous explosions in the universe, and are of great interest in astrophysics research. Despite the development of onboard trigger algorithms, significant obstacles remain in identifying weak events and effectively filtering out false detections. These challenges motivate the exploration of innovative technologies such as deep learning and quantum computing to help traditional triggers as a second verification level and increase reliability across the detection pipeline. We present a comparison of a classical and a quantum autoencoder running on simulated quantum hardware, tackling an anomaly detection task in a controlled environment. The goal is to compare their performance, with a focus on resource-constrained scenarios involving limited data and trainable parameters. Both approaches were tested using a simulated set of light curves from the anticoincidence shield of the Compton Spectrometer and Imager (COSI) soft gamma-ray telescope. Short-duration GRB signals were simulated using MEGAlib and then modified through data augmentation techniques. The results indicate that while classical autoencoders reconstruct better, achieving a Mean Squared Error (MSE) as low as 2 . 97 × 1 0 − 5 , the efficiency of the quantum autoencoders is notable with lower numbers of parameters and smaller datasets. Using only 10 trainable parameters and 100 input samples, the quantum model obtained an MSE of 2 . 24 × 1 0 − 4 , outperforming its classical counterpart (MSE of 3 . 81 × 1 0 − 2 with 705 parameters). These results show a promising role for quantum machine learning in astrophysical contexts where data related to GRBs are limited or lightweight model architectures are essential.
This study evaluates the use of Quantum Convolutional Neural Networks (QCNNs) for identifying signals resembling Gamma-Ray Bursts (GRBs) within simulated astrophysical datasets in the form of light curves. The task addressed here focuses on distinguishing GRB-like signals from background noise in simulated Cherenkov Telescope Array Observatory (CTAO) data, the next-generation astrophysical observatory for very high-energy gamma-ray science. QCNNs, a quantum counterpart of classical Convolutional Neural Networks (CNNs), leverage quantum principles to process and analyze high-dimensional data efficiently. We implemented a hybrid quantum-classical machine learning technique using the Qiskit framework, with the QCNNs trained on a quantum simulator. Several QCNN architectures were tested, employing different encoding methods such as Data Reuploading and Amplitude encoding. Key findings include that QCNNs achieved accuracy comparable to classical CNNs, often surpassing 90%, while using fewer parameters, potentially leading to more efficient models in terms of computational resources. A benchmark study further examined how hyperparameters like the number of qubits and encoding methods affected performance, with more qubits and advanced encoding methods generally enhancing accuracy but increasing complexity. QCNNs showed robust performance on time-series datasets, successfully detecting GRB signals with high precision. The research is a pioneering effort in applying QCNNs to astrophysics, offering insights into their potential and limitations. This work sets the stage for future investigations to fully realize the advantages of QCNNs in astrophysical data analysis.
Machine Learning (ML) serves as a general-purpose, highly adaptable, and versatile framework for investigating complex systems across domains. However, the resulting computational resource demands, in terms of the number of parameters and the volume of data required to train ML models, can be high, often prohibitive. This is the case in astrophysics, where multimedia space data streams usually have to be analyzed. In this context, quantum computing emerges as a compelling and promising alternative, offering the potential to address these challenges in a feasible way. Specifically, a four-step quantum machine learning (QML) workflow is proposed encompassing data encoding, quantum circuit design, model training and evaluation. Then, focusing on the data encoding step, different techniques and models are investigated within a case study centered on the Gamma-Ray Bursts (GRB) signal detection in the astrophysics domain. The results thus obtained demonstrate the effectiveness of QML in astrophysics, highlighting the critical role of data encoding, which significantly affects the QML model performance.
Reliable anomaly detection in subsea infrastructure is critical, as overlooked faults may result in severe environmental damage and substantial financial losses. Although classical machine learning approaches provide effective solutions, they often require significant computational resources and may struggle to capture the complex, high-dimensional correlations present in multivariate sensor data. In this work, we present a comprehensive evaluation of quantum and hybrid quantum–classical architectures using the 3W dataset, a large-scale industrial benchmark comprising 2,184 real and simulated subsea events. We develop an end-to-end pipeline that transforms raw multivariate time-series signals into quantum-compatible feature representations. Four hybrid architectures are systematically compared against six classical baselines in both supervised and unsupervised settings. Our results demonstrate that a compact hybrid model (Hybrid-A) achieves an F1-score of 0.9737 using only 177 trainable parameters, matching the performance of a 10,000-parameter Random Forest and exhibiting substantially higher parameter efficiency than conventional neural networks. In the unsupervised scenario, the proposed Quantum Autoencoder outperforms its classical counterpart, improving F1-score by 5.35% and ROC-AUC by 15.28% while reducing parameter count by 42%. These findings indicate that quantum-enhanced models can achieve competitive performance with significantly reduced model complexity, highlighting their potential for deployment in resource-constrained, safety-critical edge environments. Keywords: Quantum Machine Learning, Industrial Anomaly Detection, Hybrid Quantum-Classical Models, 3W Dataset, Edge Deployment, Parameter Efficiency
Anomaly detection is an important problem with applications in various domains such as fraud detection, pattern recognition or medical diagnosis. Several algorithms have been introduced using classical computing approaches. However, using quantum computing for solving anomaly detection problems in time series data is a widely unexplored research field. This paper explores the application of quantum autoencoders to time series anomaly detection. We investigate two primary techniques for classifying anomalies: (1) Analyzing the reconstruction error generated by the quantum autoencoder and (2) latent representation analysis. Our simulated experimental results, conducted across various ansaetze, demonstrate that quantum autoencoders consistently outperform classical deep learning-based autoencoders across multiple datasets. Specifically, quantum autoencoders achieve superior anomaly detection performance while utilizing 60-230 times fewer parameters and requiring five times fewer training iterations. In addition, we implement our quantum encoder on real quantum hardware. Our experimental results demonstrate that quantum autoencoders achieve anomaly detection performance on par with their simulated counterparts.
Anomaly detection on small and unbalanced datasets remains very challenging in machine learning, although this scenario is common in several domains, including healthcare, cybersecurity, finance, and energy. Data augmentation and generative AI may mitigate training-data scarcity, but they often fall short because anomalies are, by definition, unpredictable, rare, and highly diverse events compared to high-probability normal data. Overfitting to pseudo-anomalies, model collapse, high-dimensional data, uninterpretable black-box models, and validation challenges are typical issues limiting their practical applicability. In this context, quantum machine learning may provide a promising and more sustainable avenue because it can enable more interpretable models with far fewer trainable parameters and smaller datasets, implementable on energy-efficient quantum hardware. Here, we propose a novel two-step hybrid classical--quantum architecture for sequential data and test it on a realistic scenario in the global energy-transition domain, i.e., automated anomaly detection in large-scale photovoltaic plants. The achieved generalization capability and competitive prediction accuracy may pave the way for new hybrid learning models able to exploit the continuously increasing power of cloud-available and more sustainable quantum accelerators integrated with more traditional energy-hungry High Performance Computing resources.
Abstract Anomaly detection is an important problem with applications in various domains such as fraud detection, pattern recognition, or medical diagnosis. Several algorithms have been introduced using classical computing approaches. However, using quantum computing for solving anomaly detection problems in time series data is a widely unexplored research field. This paper explores the application of quantum autoencoders to time series anomaly detection. We investigate two primary techniques for classifying anomalies: (1) Analyzing the reconstruction error generated by the quantum autoencoder and (2) latent representation analysis. Our simulated experimental results, conducted across various ansaetze, demonstrate that quantum autoencoders consistently outperform classical deep learning-based autoencoders across multiple datasets. Specifically, quantum autoencoders achieve superior anomaly detection performance while utilizing 60–230 times fewer parameters and requiring five times fewer training iterations. In addition, we implement our quantum encoder on real quantum hardware. Our experimental results demonstrate that quantum autoencoders achieve anomaly detection performance on par with their simulated counterparts.