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

arXiv:2611.00885

Where Does a Variational Quantum Classifier Lose Accuracy? Angle Encoding, Readout and the Fidelity-Kernel Reference in Cyberattack Detection

Angelos Thomos, Theodore Andronikos

quant-phSubmitted Oct 31, 2026• Announced Nov 3, 2026• 0 citations
arXivHTMLPDF

Abstract

When a variational quantum classifier performs poorly at detecting network attacks, it is hard to tell which part is to blame: the encoding, the trained circuit or the readout, which turns measurements into a class. In this work, we build upon a recent amplitude-encoded classifier, and we try to separate their contributions to the NSL-KDD benchmark in noiseless simulation. We compare angle and amplitude encoding, vary one component at a time, and measure each circuit against a classical yardstick: a support-vector machine on the encoded states, whose accuracy estimates what a well-trained circuit can reach. Most of the shortfall lies in the readout. Reading one qubit, or the parity of all qubits, leaves several percentage points unused, while a readout trained together with the circuit recovers most of them from the same measurements. This also explains why some classifiers stop improving as further layers are added. The encodings suit different needs. Angle encoding gives the more accurate detector on traffic like the training data and, from eight features up, shallower circuits; amplitude encoding needs far fewer qubits and catches more attacks of types never seen in training. Finally, we report that classical models, given the same inputs, are about as accurate as the best circuits, thus, registering no quantum advantage.

AI Summary

The paper asks whether encoding, variational training, or measurement readout limits quantum-classifier accuracy in network-attack detection. It compares amplitude and angle encodings on NSL-KDD using exact noiseless simulations, varying readout and depth, with matched circuit-parameter counts for 4, 8, and 16 principal components. A fidelity-kernel SVM provides an empirical reference for linear models on the encoded density matrices; this reference is not a rigorous test-accuracy upper bound. For the four-qubit amplitude-encoded baseline, replacing single-qubit readout with a co-trained linear head raises internal accuracy from 85.6% to 94.8%, versus 97.0% for the reference SVM. In matched comparisons, global parity falls 2.6–7.4 percentage points below the corresponding reference; co-trained heads recover most of this gap using observables obtainable from the same computational-basis samples. An eight-parameter circuit with a trained head reaches 91.7%, versus 88.2% for a 72-parameter circuit with single-qubit readout. Angle encoding improves in-distribution accuracy by 1.6–4.5 points, while amplitude encoding recalls 55–61% of previously unseen attack records versus 30–45% for angle encoding. Classical kernel references exhibit the same reversal, and no quantum advantage is observed. Limitations include one legacy binary-classification benchmark, five seeds except only one for the 16-qubit angle circuit, one angle bandwidth, no device-noise simulation, and finite-shot testing only at inference. The matched comparison also changes qubit count, state-space dimension, and readout-feature count, so it does not isolate encoding from these consequences.

Little sign of AI-generated textIncrementalOverclaims

Error found: The final paragraph incorrectly describes fidelity-kernel SVM test accuracy as an estimated upper bound, although it is a regularization-dependent empirical reference rather than a bound on circuit test accuracy. The abstract's finite-shot robustness statement is broader than the evidence, which tests only seed-0 angle-encoded circuits with inference-time sampling.

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Related papers

Evaluation of Variational Quantum Classifiers (VQC) for Cyberattack Detection in the NISQ Era

Angelos Thomos, Theodore Andronikos

This paper investigates the effectiveness and structural limits of Variational Quantum Classifiers (VQC) for detecting network anomalies in the era of Noisy Intermediate-Scale Quantum (NISQ) systems. Using the official 20\% research subset of the NSL-KDD dataset, a 4-qubit classifier featuring a 24-parameter trainable ansatz was developed, utilizing amplitude encoding to embed 16 principal components. On a held-out partition of this subset, the model achieved a consistent binary-classification accuracy of 88\%. A comparative evaluation with two fundamentally different optimizers, COBYLA and SPSA, found that the observed performance plateau is not attributable to convergence to local minima, a result we interpret as consistent with an encoding-related expressibility limit rather than an optimization artifact. An architectural parity comparison with a classical neural network (a Tiny MLP with four nodes, achieving 97\% accuracy) highlighted the expressiveness gap associated with data overcompression into restricted quantum states. The model was trained on a class-balanced set spanning the 22 raw NSL-KDD attack categories and evaluated in-sample as a probe of representational capacity: under this configuration the VQC reached only 9\% accuracy and exhibited a degenerate mode collapse onto a small subset of classes. We interpret this behavior as consistent with the loss of linear separability induced by excessive compression in the quantum probability space, while explicitly noting that our experiments do not isolate the encoding from the ansatz depth, optimizer budget, and measurement-decoding scheme (see Limitations). Motivated by these observations and by Cover's theorem, we outline an alternative paradigm: a 16-qubit VQC with angle encoding that expands the Hilbert-space representation rather than relying on aggressive classical dimensionality reduction.

quant-pharXiv:2606.21715•Journal of Sensor and Actuator Networks•Jun 23, 2026•2 citations

Quantum-Optimized Feature Selection for Enhanced AI-Driven Cybersecurity

Asad Raza, Syeda Sabeeka Fatima

Machine learning Intrusion detection systems (IDS) fail with large feature spaces and imbalanced attack data. At the same time, current quantum devices are limited by noise and small qubit counts. In this work, we explore a simple hybrid approach in which quantum circuits are used only for feature transformation, while a classical model performs the final classification. Normalized NSL-KDD traffic records are first reduced by principal component analysis (PCA), then encoded into quantum states and processed by a low-depth variational quantum circuit (VQC). The measured outputs form a compact feature vector that is fed to a support vector machine (SVM). On the NSL-KDD dataset, the proposed VQC+SVM model reaches $80.9 \%$ accuracy and an F1-score of 0.814, which is higher than PCA+SVM ($\mathbf{7 3 . 0 \%}$) and autoencoder+SVM ($\mathbf{7 5 . 7 \%}$). An ablation study over different circuit depths and qubit counts shows that a 4-qubit, 3-layer circuit provides the most stable performance under NISQ-like conditions.

International Multi-Topic ICT Conference (IMTIC)•Dec 29, 2025•0 citations

Comparative study of amplitude versus angle encoding in gate-based quantum machine learning

Shashank Kumar, Ashok Kumar

We investigate two fundamental data encoding strategies for gate-based quantum computers: amplitude encoding and angle encoding. In amplitude encoding, a normalized -dimensional classical vector is mapped into the probability amplitudes of a -qubit quantum state, whereas angle encoding uses one or more single-qubit rotation gates per feature to encode data as angles.[1, 2] We construct a toy binary classification dataset with four real-valued features and compare these encoding schemes using a variational quantum classifier (VQC) approach. Our results indicate that angle encoding often yields slightly higher classification accuracy (e.g. 82% vs. 75% in one experiment) at the cost of requiring more qubits, while amplitude encoding compresses features into fewer qubits (2 vs. 4) but demands deeper state-preparation circuits. These findings align with previous studies that report higher accuracy for rotation encodings (e.g. vs. in MNIST experiments [3]) and note that the encoding choice acts as a hyperparameter affecting VQC performance. [4]

International Journal of Physics and Mathematics•Jul 1, 2025•2 citations

Variational Quantum Algorithms for Image Classification

Andrej Sum-Shik, Peter Thoma

Abstract Variational quantum algorithms are among the most widely studied approaches for applying near-term quantum devices to learning tasks. In this work we investigate a supervised quantum-autoencoder-style classifier that encodes images by amplitude encoding and predicts labels from a designated readout sub-register. Building on the quantum-autoencoder (QAE)-style classifier of Asaoka and Kudo, we make three contributions. First, we give a rigorous equivalence result showing that, for computational-basis label states, the original SWAP-test-based training objective can be estimated without ancilla and reference registers by directly measuring the readout subsystem. This simplification reduces circuit resources while preserving the same training signal. Second, we propose a lightweight hybrid extension: instead of using only the probability of the correct readout bitstring, we estimate the full readout distribution and apply a learnable temperature-and-bias post-processing followed by a softmax normalization. Across MNIST, KMNIST, and Fashion-MNIST (restricted to eight classes), this classical post-processing yields substantial and statistically significant improvements in evaluation accuracy, consistently across cross-entropy and mean-squared-error objectives. Finally, to interpret these gains we analyze Fisher-geometry summaries based on effective dimension and complementary Fisher-spectrum diagnostics for both the quantum models and the classical baselines. Empirically, the hybrid quantum model is associated with a more spectrally concentrated and anisotropic Fisher geometry, with Fisher mass concentrated in fewer dominant directions and with a substantially larger local Fisher scale after training.

•Apr 14, 2026•0 citations

Rigorous Benchmarking of Variational Quantum Classifiers for Cancer Classification: Encoding Bandwidth and a Matched, Tuned Repeated Cross-Validation Study

B RAVI KIRAN PATNAIK

Variational quantum classifiers (VQCs) are often compared with classical models under unequal data partitions, tuning budgets, or statistical treatment. We benchmark a 4-qubit VQC against logistic regression, an RBF-SVM, random forest, a quantum-kernel SVM (QSVM), and a majority-class control on Breast Cancer Wisconsin (n=569), Coimbra (n=116), and Golub leukemia (n=72), using five stratified folds across five seeds (25 paired blocks per dataset), identical training-fold-fitted preprocessing, and Nadeau–Bengio corrected paired tests with Holm correction. A first arm with a frozen configuration and the common [0, π] angle encoding left the VQC at 0.563, 0.515, and 0.576 accuracy (majority 0.617, 0.552, 0.652). In a second arm, in which every model was tuned by the same inner cross-validation and the encoding scale was a tuned hyperparameter, VQC accuracy rose to 0.864, 0.616, and 0.747. On Wisconsin the VQC stayed significantly below every matched classical model (differences 0.079–0.097, Holm p≤0.006) but above the majority class; on Golub it was significantly below only the SVM (Holm p=0.016); on Coimbra no comparison was significant. A QSVM with the same feature map exceeded the VQC numerically on every dataset, and three optimizers reached the same training loss in pilot diagnostics, pointing to the encoding and the variational head rather than the optimizer. The Wisconsin arms also differ in sample size (120 versus 569). Results apply to this low-qubit, PCA-compressed, exactly simulated setting and do not indicate a quantum advantage.

•Oct 1, 2026•0 citations

Evaluating Angle and Amplitude Encoding Strategies for Variational Quantum Machine Learning: Their Impact on Model's Accuracy

Antonio Tudisco, Andrea Marchesin, Maurizio Zamboni +2 more

ABSTRACT Recent advancements in Quantum Computing and Machine Learning have increased attention to Quantum Machine Learning (QML), which aims to develop machine learning models by exploiting the quantum computing paradigm. One of the widely used models in this area is the Variational Quantum Circuit (VQC), a hybrid model where the quantum circuit handles data inference while classical optimization adjusts the parameters of the circuit. The quantum circuit consists of an encoding layer, which loads data into the circuit, and a template circuit, known as the ansatz, responsible for processing the data. This work involves performing an analysis by considering both Amplitude‐ and Angle‐encoding models, and examining how the type of rotational gate applied affects the classification performance of the model. This comparison is carried out by training the different models on two datasets, Wine and Diabetes, and evaluating their performance. The study demonstrates that, under identical model topologies, the difference in accuracy between the best and worst models ranges from 10% to 30%, with differences reaching up to 41%. Moreover, the results highlight how the choice of rotational gates used in encoding can significantly impact the model's classification performance. The findings confirm that the embedding represents a hyperparameter for VQC models.

arXiv:2508.00768•Advanced Quantum Technologies•Apr 1, 2026•12 citations
When a variational quantum classifier performs poorly at detecting network attacks, it is hard to tell which part is to blame: the encoding, the trained circuit or the readout, which turns measurements into a class. In this work, we build upon a recent amplitude-encoded classifier, and we try to separate their contributions to the NSL-KDD benchmark in noiseless simulation. We compare angle and amplitude encoding, vary one component at a time, and measure each circuit against a classical yardstick: a support-vector machine on the encoded states, whose accuracy estimates what a well-trained circuit can reach. Most of the shortfall lies in the readout. Reading one qubit, or the parity of all qubits, leaves several percentage points unused, while a readout trained together with the circuit recovers most of them from the same measurements. This also explains why some classifiers stop improving as further layers are added. The encodings suit different needs. Angle encoding gives the more accurate detector on traffic like the training data and, from eight features up, shallower circuits; amplitude encoding needs far fewer qubits and catches more attacks of types never seen in training. Finally, we report that classical models, given the same inputs, are about as accurate as the best circuits, thus, registering no quantum advantage.
The paper asks whether encoding, variational training, or measurement readout limits quantum-classifier accuracy in network-attack detection. It compares amplitude and angle encodings on NSL-KDD using exact noiseless simulations, varying readout and depth, with matched circuit-parameter counts for 4, 8, and 16 principal components. A fidelity-kernel SVM provides an empirical reference for linear models on the encoded density matrices; this reference is not a rigorous test-accuracy upper bound.
For the four-qubit amplitude-encoded baseline, replacing single-qubit readout with a co-trained linear head raises internal accuracy from 85.6% to 94.8%, versus 97.0% for the reference SVM. In matched comparisons, global parity falls 2.6–7.4 percentage points below the corresponding reference; co-trained heads recover most of this gap using observables obtainable from the same computational-basis samples. An eight-parameter circuit with a trained head reaches 91.7%, versus 88.2% for a 72-parameter circuit with single-qubit readout. Angle encoding improves in-distribution accuracy by 1.6–4.5 points, while amplitude encoding recalls 55–61% of previously unseen attack records versus 30–45% for angle encoding. Classical kernel references exhibit the same reversal, and no quantum advantage is observed.
Limitations include one legacy binary-classification benchmark, five seeds except only one for the 16-qubit angle circuit, one angle bandwidth, no device-noise simulation, and finite-shot testing only at inference. The matched comparison also changes qubit count, state-space dimension, and readout-feature count, so it does not isolate encoding from these consequences.
This paper investigates the effectiveness and structural limits of Variational Quantum Classifiers (VQC) for detecting network anomalies in the era of Noisy Intermediate-Scale Quantum (NISQ) systems. Using the official 20\% research subset of the NSL-KDD dataset, a 4-qubit classifier featuring a 24-parameter trainable ansatz was developed, utilizing amplitude encoding to embed 16 principal components. On a held-out partition of this subset, the model achieved a consistent binary-classification accuracy of 88\%. A comparative evaluation with two fundamentally different optimizers, COBYLA and SPSA, found that the observed performance plateau is not attributable to convergence to local minima, a result we interpret as consistent with an encoding-related expressibility limit rather than an optimization artifact. An architectural parity comparison with a classical neural network (a Tiny MLP with four nodes, achieving 97\% accuracy) highlighted the expressiveness gap associated with data overcompression into restricted quantum states. The model was trained on a class-balanced set spanning the 22 raw NSL-KDD attack categories and evaluated in-sample as a probe of representational capacity: under this configuration the VQC reached only 9\% accuracy and exhibited a degenerate mode collapse onto a small subset of classes. We interpret this behavior as consistent with the loss of linear separability induced by excessive compression in the quantum probability space, while explicitly noting that our experiments do not isolate the encoding from the ansatz depth, optimizer budget, and measurement-decoding scheme (see Limitations). Motivated by these observations and by Cover's theorem, we outline an alternative paradigm: a 16-qubit VQC with angle encoding that expands the Hilbert-space representation rather than relying on aggressive classical dimensionality reduction.
Machine learning Intrusion detection systems (IDS) fail with large feature spaces and imbalanced attack data. At the same time, current quantum devices are limited by noise and small qubit counts. In this work, we explore a simple hybrid approach in which quantum circuits are used only for feature transformation, while a classical model performs the final classification. Normalized NSL-KDD traffic records are first reduced by principal component analysis (PCA), then encoded into quantum states and processed by a low-depth variational quantum circuit (VQC). The measured outputs form a compact feature vector that is fed to a support vector machine (SVM). On the NSL-KDD dataset, the proposed VQC+SVM model reaches 80.9%80.9 \%80.9% accuracy and an F1-score of 0.814, which is higher than PCA+SVM (73.0%\mathbf{7 3 . 0 \%}73.0%) and autoencoder+SVM (75.7%\mathbf{7 5 . 7 \%}75.7%). An ablation study over different circuit depths and qubit counts shows that a 4-qubit, 3-layer circuit provides the most stable performance under NISQ-like conditions.
We investigate two fundamental data encoding strategies for gate-based quantum computers: amplitude encoding and angle encoding. In amplitude encoding, a normalized -dimensional classical vector is mapped into the probability amplitudes of a -qubit quantum state, whereas angle encoding uses one or more single-qubit rotation gates per feature to encode data as angles.[1, 2] We construct a toy binary classification dataset with four real-valued features and compare these encoding schemes using a variational quantum classifier (VQC) approach. Our results indicate that angle encoding often yields slightly higher classification accuracy (e.g. 82% vs. 75% in one experiment) at the cost of requiring more qubits, while amplitude encoding compresses features into fewer qubits (2 vs. 4) but demands deeper state-preparation circuits. These findings align with previous studies that report higher accuracy for rotation encodings (e.g. vs. in MNIST experiments [3]) and note that the encoding choice acts as a hyperparameter affecting VQC performance. [4]
Abstract Variational quantum algorithms are among the most widely studied approaches for applying near-term quantum devices to learning tasks. In this work we investigate a supervised quantum-autoencoder-style classifier that encodes images by amplitude encoding and predicts labels from a designated readout sub-register. Building on the quantum-autoencoder (QAE)-style classifier of Asaoka and Kudo, we make three contributions. First, we give a rigorous equivalence result showing that, for computational-basis label states, the original SWAP-test-based training objective can be estimated without ancilla and reference registers by directly measuring the readout subsystem. This simplification reduces circuit resources while preserving the same training signal. Second, we propose a lightweight hybrid extension: instead of using only the probability of the correct readout bitstring, we estimate the full readout distribution and apply a learnable temperature-and-bias post-processing followed by a softmax normalization. Across MNIST, KMNIST, and Fashion-MNIST (restricted to eight classes), this classical post-processing yields substantial and statistically significant improvements in evaluation accuracy, consistently across cross-entropy and mean-squared-error objectives. Finally, to interpret these gains we analyze Fisher-geometry summaries based on effective dimension and complementary Fisher-spectrum diagnostics for both the quantum models and the classical baselines. Empirically, the hybrid quantum model is associated with a more spectrally concentrated and anisotropic Fisher geometry, with Fisher mass concentrated in fewer dominant directions and with a substantially larger local Fisher scale after training.
Variational quantum classifiers (VQCs) are often compared with classical models under unequal data partitions, tuning budgets, or statistical treatment. We benchmark a 4-qubit VQC against logistic regression, an RBF-SVM, random forest, a quantum-kernel SVM (QSVM), and a majority-class control on Breast Cancer Wisconsin (n=569), Coimbra (n=116), and Golub leukemia (n=72), using five stratified folds across five seeds (25 paired blocks per dataset), identical training-fold-fitted preprocessing, and Nadeau–Bengio corrected paired tests with Holm correction. A first arm with a frozen configuration and the common [0, π] angle encoding left the VQC at 0.563, 0.515, and 0.576 accuracy (majority 0.617, 0.552, 0.652). In a second arm, in which every model was tuned by the same inner cross-validation and the encoding scale was a tuned hyperparameter, VQC accuracy rose to 0.864, 0.616, and 0.747. On Wisconsin the VQC stayed significantly below every matched classical model (differences 0.079–0.097, Holm p≤0.006) but above the majority class; on Golub it was significantly below only the SVM (Holm p=0.016); on Coimbra no comparison was significant. A QSVM with the same feature map exceeded the VQC numerically on every dataset, and three optimizers reached the same training loss in pilot diagnostics, pointing to the encoding and the variational head rather than the optimizer. The Wisconsin arms also differ in sample size (120 versus 569). Results apply to this low-qubit, PCA-compressed, exactly simulated setting and do not indicate a quantum advantage.
ABSTRACT Recent advancements in Quantum Computing and Machine Learning have increased attention to Quantum Machine Learning (QML), which aims to develop machine learning models by exploiting the quantum computing paradigm. One of the widely used models in this area is the Variational Quantum Circuit (VQC), a hybrid model where the quantum circuit handles data inference while classical optimization adjusts the parameters of the circuit. The quantum circuit consists of an encoding layer, which loads data into the circuit, and a template circuit, known as the ansatz, responsible for processing the data. This work involves performing an analysis by considering both Amplitude‐ and Angle‐encoding models, and examining how the type of rotational gate applied affects the classification performance of the model. This comparison is carried out by training the different models on two datasets, Wine and Diabetes, and evaluating their performance. The study demonstrates that, under identical model topologies, the difference in accuracy between the best and worst models ranges from 10% to 30%, with differences reaching up to 41%. Moreover, the results highlight how the choice of rotational gates used in encoding can significantly impact the model's classification performance. The findings confirm that the embedding represents a hyperparameter for VQC models.