Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS)

Quantum-Inspired Neural Networks vs Machine Learning and Deep Learning: Comprehensive Benchmarking for Malicious QR Code and URL Detection

Hemalatha Raavi, Deepa Manasa Yerra, Chinta Poojitha, D. V. R. S. Aditya, Gottumukkala Dheeraj Varma, Dandangi Rufus Raj

Published May 28, 2026• 0 citations
Publisher

Abstract

Quantum-inspired machine learning leverages quantum computing principles on classical hardware for enhanced pattern recognition in cybersecurity applications. This paper presents quantum-inspired neural network architectures for malicious QR code detection and URL threat classification with comprehensive benchmarking against$\mathbf{1 5} \boldsymbol{+}$traditional ML and DL approaches. The proposed quantum-inspired CNN incorporates interference patterns through dual-activation pathways combining ReLU and tanh functions, while the quantum-inspired DNN employs triple interference mechanisms with attention- like gating and multi-path entanglement. Experimental evaluation on$20,000 \text{QR}$code images demonstrates 100 % classification accuracy compared to Random Forest (87.3 %), SVM (89.1 %), conventional CNN (95.2 %), and Inception-v3$(96.8 \%)$. URL classification across 20,000 samples spanning four threat categories achieves 94.27 % accuracy, outperforming Random Forest (84.6%), SVM (82.1%), conventional DNN (89.3 %), LSTM (91.2 %), and Transformer (92.8 %). The quantum-inspired architectures demonstrate efficient convergence with 28 % faster training while maintaining computational overhead within$25-30 \%$. Results validate quantum-inspired approaches as state-of- the-art cybersecurity solutions deployable on conventional GPU infrastructure.

DOI
10.1109/icssas68835.2026.11559511

Related papers

Quantum-Enhanced Machine Learning for Cybersecurity: Evaluating Malicious URL Detection

Lauren Eze, Umair B. Chaudhry, Hamid Jahankhani

The constant rise of malicious URLs continues to pose significant threats and challenges in cybersecurity, with attackers increasingly evading classical detection methods like blacklists and heuristic-based systems. While machine learning (ML) techniques such as SVMs and CNNs have improved detection, their accuracy and scalability remain limited for emerging adversarial approaches. Quantum machine learning (QML) is a transformative strategy that relies on quantum computation and high-dimensional feature spaces to potentially overcome classical computational limitations. However, the accuracy of QML models such as QSVM and QCNN for URL detection in comparison to classical models remains unexplored. This study evaluates ML models (SVMs and CNNs) and QML models (QSVMs and QCNNs) on a dataset, employing data preprocessing techniques such as outliers, feature scaling and feature selection with ANOVA and PCA. Quantum models utilized ZZFeatureMap and ZFeatureMap for data encoding, to transfer original data to qubits. The achieved results showed that CNNs outperformed QCNNs and QSVMs outperformed SVMs in the performance evaluation, demonstrating a competitive potential of quantum computing. QML shows promise for cybersecurity, particularly given the QSVM’s kernel advantages, but current hardware limits the QCNN’s practicality. The significance of this research is to contribute to the growing body of knowledge in cybersecurity by providing a comparative analysis of classical and quantum ML models for classifying malicious URLs.

Electronics•Apr 29, 2025•31 citations

Do We Really Need Quantum Machine Learning?: A Multidimensional Empirical Study

Sudip Vhaduri, Ryan Gammon, Sayanton Dibbo

The rapid growth of computer vision and increasingly complex image recognition tasks has exposed fundamental computational limitations of classical machine learning models, motivating the exploration of quantum computing as an emerging new paradigm. This paper presents a comprehensive benchmarking study of classical and quantum machine learning models for image recognition on the MNIST handwritten digit dataset, evaluating both traditional models, a Classical Support Vector Machine (CSVM) and a Quantum Support Vector Machine (QSVM), and deep neural network models, a Classical Convolutional Neural Network (CCNN) and a Quantum Convolutional Neural Network (QCNN), across four performance dimensions: classification accuracy, computational runtime, parameter count, and memory requirements. Experiments are conducted as functions of both feature dimensionality and sample size, and across CPU and GPU execution environments, providing a controlled, multidimensional comparison to address gaps in prior work. For the SVM-based models, QSVM consistently outperforms CSVM in accuracy, reaching $\sim$ 0.90 versus $\sim$ 0.85 at 1,000 samples, with a higher computational cost. A feature count of 10 qubits and a sample size in the range of 200 -- 500 emerge as practical operating points that balance accuracy and runtime. For the neural network models, CCNN and QCNN achieve comparable classification accuracy, both exceeding 0.96 at 64 features and 60,000 samples, yet QCNN offers superior parameter and memory efficiency at higher feature counts, while incurring higher runtime. Across both model families, quantum models consistently outperform classical models by greater margins in accuracy as feature dimensionality or sample size increases.

cs.CVCross-listedarXiv:2605.27923•May 28, 2026•4 citations

Comparative Analysis of Conventional and Quantum Machine Learning for Cybersecurity

Nandala Narayanamma, Karnam Madhavi

Cyberspace menace has grown to a more advanced level where it is very challenging to counter it using traditional defense systems. Although classical approaches to machine learning (ML) have been extensively used in intrusion and anomaly detection, Support Vector Machines (SVM) and Neural Networks (NN) are often limited in one way or another by when handling high-dimensional, high complexity, and actively changing cyber-attack patterns. According to the laws of superposition and entanglement, Quantum Machine Learning (QML) is a potential alternative that can increase the efficiency of calculations and reach a higher detection rate. We conduct a comparative analysis of the traditional ML models (SVM, NN) with quantum ML models (Quantum SVM and Quantum Neural Networks) on benchmark datasets, such as the NSL -KDD, KDD Cup 1999 and a malware detection dataset in this paper. These findings have shown that the QML models, especially the QSVM, are better than the classical models in terms of accuracy (around 94.8) and trainings (around 85 seconds). The level of these improvements is supported with statistical tests. However, the results were obtained in an idealized simulation case and they are limited to the limitations of NISQ hardware in the field. The paper concludes by positively identifying that QML represents an exciting and efficient prospect to next-generation cybersecurity systems though its implementation to date is restricted due to the existing quantum hardware resources.

•Aug 1, 2026•0 citations

Quantum-Driven Machine Learning Framework for Enhanced Detection of Malicious URLs in Cybersecurity

S V Bhaskar, K R Spoorthi, C M Harshith +3 more

As cyber threats grow in sophistication, so must our detection methods. Traditional machine learning models are generally effective against simpler cyber behaviors but suffer from inefficiencies in dealing with high-dimensional data and being compromised by adversarial attacks. This paper explores the application of Quantum-Enhanced Machine Learning (QEML) models in the area of cybersecurity, specifically for malicious URL detection. We propose a hybrid quantum-classical technology leveraging the quantum computer’s unique capability to efficiently search a high-dimensional feature space. The proposed quantum-enhanced model was benchmarked against a classical project; on significant portions of the dataset, the quantum model proved to be vastly more accurate and robust when assessed against the classical techniques.

International Conference on Sustainable Computing and Data Communication Systems (ICSCDS)•Aug 6, 2025•1 citation

Quantum Machine Learning in Cybersecurity: A Hybrid Convolutional Neural Network with Skip Connections for QR Code Phishing Detection

Md. Farhan Shahriyar

Abstract QR code phishing, commonly called quishing, has become a significant cybersecurity threat that exploits the widespread adoption of QR codes in contactless payment, ticketing and authentication. Existing countermeasures rely almost entirely on decoding the code and screening the extracted URL, which fails when the payload is obfuscated, when the destination is a zero-day domain or when no network connection is available at scan time. This study proposes a Hybrid Quantum-Classical Convolutional Neural Network (HQCNN) that classifies QR code images directly from their visual structure without any decoding step. The architecture couples a classical convolutional backbone with a parameterised variational quantum circuit using angle embedding, trainable rotations and a configurable entangling layer. A further variant, HybridSkip, introduces a residual bypass carrying backbone features directly to the classification head alongside the quantum path which preserves gradient flow, retains spatial context and removes the wide compensatory layer the standard hybrid model needs. Six configurations are evaluated on a balanced corpus of QR code images under identical stratified five-fold cross-validation splits and a held-out test set. Hybrid Skip with 6 qubits and 4 variational layers reaches an F1-score 3.3 percent higher and a recall 6.8 percent higher than the strongest classical baseline while using 37.6 percent fewer trainable parameters. Paired tests across folds confirm the recall gain is statistically significant with a large effect size. The results support parameter-efficient hybrid architectures as a route to proactive quishing defence on resource constrained endpoints in the noisy intermediate scale quantum era.

•Sep 29, 2026•0 citations

Quantum deep convolutional neural networks for damage classification

Abu Bakar Siddique, Young‐Jin Cha

Quantum computing, based on fundamental quantum-mechanical principles and phenomena such as superposition, entanglement, and interference, offers access to high-dimensional quantum feature spaces and complex correlations that may be difficult to efficiently represent using classical computing, along with potential computational advantages through quantum hardware. Therefore, quantum computing has emerged as a promising paradigm for addressing limitations of classical deep learning (DL) through fundamentally different computational mechanisms. Quantum computing-based DL approaches have been applied to image classification; however, most rely on existing quantum computing libraries that impose substantial computational costs, limiting their broader applicability. In this study, hybrid quantum convolutional neural networks (HQCNNs) are proposed for structural damage classification without relying on existing quantum computing libraries. The proposed architectures integrate quanvolutional layers, composed of encoding, variational quantum circuits, and measurement, with conventional convolutional layers in a CNN architecture, enabling quantum-enhanced feature extraction while maintaining compatibility with classical computing systems. The scratch-implemented HQCNN (HQCNN(S)), its library-based counterpart (HQCNN(P)), and the classical CNN are compared in terms of test accuracy, computational cost, and training efficiency. Experimental results demonstrate that HQCNN(S) ( at ) achieves up to 176.57 faster computation than HQCNN(P) and 3.1% higher test accuracy than the classical CNN, while reducing the number of trainable parameters by 60.2%. Moreover, HQCNN(S) achieves comparable performance to the best-performing classical CNN using 75% less training data.

Computer-Aided Civil and Infrastructure Engineering•Sep 10, 2026•0 citations
Quantum-inspired machine learning leverages quantum computing principles on classical hardware for enhanced pattern recognition in cybersecurity applications. This paper presents quantum-inspired neural network architectures for malicious QR code detection and URL threat classification with comprehensive benchmarking against15+\mathbf{1 5} \boldsymbol{+}15+traditional ML and DL approaches. The proposed quantum-inspired CNN incorporates interference patterns through dual-activation pathways combining ReLU and tanh functions, while the quantum-inspired DNN employs triple interference mechanisms with attention- like gating and multi-path entanglement. Experimental evaluation on20,000QR20,000 \text{QR}20,000QRcode images demonstrates 100 % classification accuracy compared to Random Forest (87.3 %), SVM (89.1 %), conventional CNN (95.2 %), and Inception-v3(96.8%)(96.8 \%)(96.8%). URL classification across 20,000 samples spanning four threat categories achieves 94.27 % accuracy, outperforming Random Forest (84.6%), SVM (82.1%), conventional DNN (89.3 %), LSTM (91.2 %), and Transformer (92.8 %). The quantum-inspired architectures demonstrate efficient convergence with 28 % faster training while maintaining computational overhead within25−30%25-30 \%25−30%. Results validate quantum-inspired approaches as state-of- the-art cybersecurity solutions deployable on conventional GPU infrastructure.
The constant rise of malicious URLs continues to pose significant threats and challenges in cybersecurity, with attackers increasingly evading classical detection methods like blacklists and heuristic-based systems. While machine learning (ML) techniques such as SVMs and CNNs have improved detection, their accuracy and scalability remain limited for emerging adversarial approaches. Quantum machine learning (QML) is a transformative strategy that relies on quantum computation and high-dimensional feature spaces to potentially overcome classical computational limitations. However, the accuracy of QML models such as QSVM and QCNN for URL detection in comparison to classical models remains unexplored. This study evaluates ML models (SVMs and CNNs) and QML models (QSVMs and QCNNs) on a dataset, employing data preprocessing techniques such as outliers, feature scaling and feature selection with ANOVA and PCA. Quantum models utilized ZZFeatureMap and ZFeatureMap for data encoding, to transfer original data to qubits. The achieved results showed that CNNs outperformed QCNNs and QSVMs outperformed SVMs in the performance evaluation, demonstrating a competitive potential of quantum computing. QML shows promise for cybersecurity, particularly given the QSVM’s kernel advantages, but current hardware limits the QCNN’s practicality. The significance of this research is to contribute to the growing body of knowledge in cybersecurity by providing a comparative analysis of classical and quantum ML models for classifying malicious URLs.
The rapid growth of computer vision and increasingly complex image recognition tasks has exposed fundamental computational limitations of classical machine learning models, motivating the exploration of quantum computing as an emerging new paradigm. This paper presents a comprehensive benchmarking study of classical and quantum machine learning models for image recognition on the MNIST handwritten digit dataset, evaluating both traditional models, a Classical Support Vector Machine (CSVM) and a Quantum Support Vector Machine (QSVM), and deep neural network models, a Classical Convolutional Neural Network (CCNN) and a Quantum Convolutional Neural Network (QCNN), across four performance dimensions: classification accuracy, computational runtime, parameter count, and memory requirements. Experiments are conducted as functions of both feature dimensionality and sample size, and across CPU and GPU execution environments, providing a controlled, multidimensional comparison to address gaps in prior work. For the SVM-based models, QSVM consistently outperforms CSVM in accuracy, reaching ∼\sim∼ 0.90 versus ∼\sim∼ 0.85 at 1,000 samples, with a higher computational cost. A feature count of 10 qubits and a sample size in the range of 200 -- 500 emerge as practical operating points that balance accuracy and runtime. For the neural network models, CCNN and QCNN achieve comparable classification accuracy, both exceeding 0.96 at 64 features and 60,000 samples, yet QCNN offers superior parameter and memory efficiency at higher feature counts, while incurring higher runtime. Across both model families, quantum models consistently outperform classical models by greater margins in accuracy as feature dimensionality or sample size increases.
Cyberspace menace has grown to a more advanced level where it is very challenging to counter it using traditional defense systems. Although classical approaches to machine learning (ML) have been extensively used in intrusion and anomaly detection, Support Vector Machines (SVM) and Neural Networks (NN) are often limited in one way or another by when handling high-dimensional, high complexity, and actively changing cyber-attack patterns. According to the laws of superposition and entanglement, Quantum Machine Learning (QML) is a potential alternative that can increase the efficiency of calculations and reach a higher detection rate. We conduct a comparative analysis of the traditional ML models (SVM, NN) with quantum ML models (Quantum SVM and Quantum Neural Networks) on benchmark datasets, such as the NSL -KDD, KDD Cup 1999 and a malware detection dataset in this paper. These findings have shown that the QML models, especially the QSVM, are better than the classical models in terms of accuracy (around 94.8) and trainings (around 85 seconds). The level of these improvements is supported with statistical tests. However, the results were obtained in an idealized simulation case and they are limited to the limitations of NISQ hardware in the field. The paper concludes by positively identifying that QML represents an exciting and efficient prospect to next-generation cybersecurity systems though its implementation to date is restricted due to the existing quantum hardware resources.
As cyber threats grow in sophistication, so must our detection methods. Traditional machine learning models are generally effective against simpler cyber behaviors but suffer from inefficiencies in dealing with high-dimensional data and being compromised by adversarial attacks. This paper explores the application of Quantum-Enhanced Machine Learning (QEML) models in the area of cybersecurity, specifically for malicious URL detection. We propose a hybrid quantum-classical technology leveraging the quantum computer’s unique capability to efficiently search a high-dimensional feature space. The proposed quantum-enhanced model was benchmarked against a classical project; on significant portions of the dataset, the quantum model proved to be vastly more accurate and robust when assessed against the classical techniques.
Abstract QR code phishing, commonly called quishing, has become a significant cybersecurity threat that exploits the widespread adoption of QR codes in contactless payment, ticketing and authentication. Existing countermeasures rely almost entirely on decoding the code and screening the extracted URL, which fails when the payload is obfuscated, when the destination is a zero-day domain or when no network connection is available at scan time. This study proposes a Hybrid Quantum-Classical Convolutional Neural Network (HQCNN) that classifies QR code images directly from their visual structure without any decoding step. The architecture couples a classical convolutional backbone with a parameterised variational quantum circuit using angle embedding, trainable rotations and a configurable entangling layer. A further variant, HybridSkip, introduces a residual bypass carrying backbone features directly to the classification head alongside the quantum path which preserves gradient flow, retains spatial context and removes the wide compensatory layer the standard hybrid model needs. Six configurations are evaluated on a balanced corpus of QR code images under identical stratified five-fold cross-validation splits and a held-out test set. Hybrid Skip with 6 qubits and 4 variational layers reaches an F1-score 3.3 percent higher and a recall 6.8 percent higher than the strongest classical baseline while using 37.6 percent fewer trainable parameters. Paired tests across folds confirm the recall gain is statistically significant with a large effect size. The results support parameter-efficient hybrid architectures as a route to proactive quishing defence on resource constrained endpoints in the noisy intermediate scale quantum era.
Quantum computing, based on fundamental quantum-mechanical principles and phenomena such as superposition, entanglement, and interference, offers access to high-dimensional quantum feature spaces and complex correlations that may be difficult to efficiently represent using classical computing, along with potential computational advantages through quantum hardware. Therefore, quantum computing has emerged as a promising paradigm for addressing limitations of classical deep learning (DL) through fundamentally different computational mechanisms. Quantum computing-based DL approaches have been applied to image classification; however, most rely on existing quantum computing libraries that impose substantial computational costs, limiting their broader applicability. In this study, hybrid quantum convolutional neural networks (HQCNNs) are proposed for structural damage classification without relying on existing quantum computing libraries. The proposed architectures integrate quanvolutional layers, composed of encoding, variational quantum circuits, and measurement, with conventional convolutional layers in a CNN architecture, enabling quantum-enhanced feature extraction while maintaining compatibility with classical computing systems. The scratch-implemented HQCNN (HQCNN(S)), its library-based counterpart (HQCNN(P)), and the classical CNN are compared in terms of test accuracy, computational cost, and training efficiency. Experimental results demonstrate that HQCNN(S) ( at ) achieves up to 176.57 faster computation than HQCNN(P) and 3.1% higher test accuracy than the classical CNN, while reducing the number of trainable parameters by 60.2%. Moreover, HQCNN(S) achieves comparable performance to the best-performing classical CNN using 75% less training data.