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:2610.23692

Expressivity of IQP Feature-Map Quantum Kernels via Reproducing Kernel Hilbert Spaces

Valeria Repetto, Stefano Melchionna, Romina D'Aurizio

quant-phcs.LGSubmitted Oct 16, 2026• Announced Oct 19, 2026• 0 citations
arXivHTMLPDF

Abstract

Quantum kernels are among the most promising methods in quantum machine learning, providing a natural way to embed data into high-dimensional Hilbert spaces via quantum circuits and capturing non-linear relationships in data. Their learning behaviour is governed by the function class they represent, which is formally characterized by the structure of the associated Reproducing Kernel Hilbert Space (RKHS). However, a theoretical characterization of this function class, and of how it depends on the choice of feature map and its hyperparameters, is still lacking. In this work, we focus on the Instantaneous Quantum Polynomial (IQP) feature map and derive a Fourier representation of the associated RKHS generators as finite sums of trigonometric functions evaluated on polynomially transformed input variables, with a finite set of accessible frequencies determined by the circuit construction and scaled by the bandwidth hyperparameter. This representation identifies three bandwidth regimes with qualitatively distinct expressive properties from localised, through multimodal, to periodic. Numerical experiments on synthetic datasets with controlled periodic and multimodal structure, across data dimensions and feature-map configurations, reproduce the three regimes in both supervised classification and unsupervised anomaly detection, benchmarked against a classical Gaussian kernel. These results make the relationship between expressivity and data structure explicit for IQP kernels: expressive power translates into a practical benefit only when the accessible frequencies match with the structure of the data, which provides a criterion for selecting the bandwidth when the periodic or multimodal structure of data is known.

AI Summary

The paper asks how the function class represented by an IQP fidelity kernel depends on its feature map and hyperparameters, particularly input bandwidth $b$. It studies this through the generators $k(x,\cdot)$ of the associated reproducing kernel Hilbert space (RKHS). Extending an existing circuit Fourier-expansion framework, it represents the generators as finite trigonometric sums evaluated on polynomial transformations of the inputs of degree at most two. For Pauli-$Z$ encodings, elementary frequency differences lie in $\{-2,0,2\}$; repeated encodings permit finite sums of these differences. The representation motivates three qualitative bandwidth regimes: Gaussian-like/local at small $b$, multimodal at intermediate $b$, and strongly oscillatory at large $b$. Polynomial coordinate transformations mean the kernels are generally not exactly periodic in the original inputs. Numerical tests use synthetic periodic and Gaussian-mixture datasets, SVM classification and one-class SVM anomaly detection, with $D=2,4,6$ qubits, one to three circuit repetitions, and linear or full entanglement. On periodic classification data, a representative two-qubit configuration achieves F1 scores of $0.75$–$0.84$ when $b$ matches the label frequency, versus $0.46$–$0.72$ otherwise. On multimodal data, performance typically peaks around $b=1$–$1.5$; quantum and Gaussian-kernel performance ranges largely overlap. The stated limitations are that numerical evidence is restricted to controlled synthetic datasets and the theoretical characterization concerns IQP feature maps. Applicability to real-world data and other feature-map families remains open.

Little sign of AI-generated textIncrementalPossible errorMissing comparisonOverclaims

Error found: The main theorem does not establish the fidelity-kernel formula used elsewhere because its kernel definition omits both the adjoint and the absolute square. The theorem's scalar-coefficient expansion incorrectly forces a rank-one kernel because the proof drops the state’s output basis index and mishandles tensor-product indices and phases. The stated pair encoding produces both b-dependent linear terms and b^2-dependent quadratic terms, contradicting a uniform linear bandwidth rescaling of frequencies in the original inputs. Changing linear to full entanglement changes data-dependent pair encodings, so it cannot generally be described as changing only expansion coefficients while preserving accessible frequency support.

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

Comments
19 pages, 11 Figures , Supplementary

Related papers

Impact of Quantum Feature Maps on the Geometry of Induced Kernels in Quantum Kernel Methods

Łukasz Pieczke

This dataset contains the complete experimental results and supporting data generated in the study of quantum feature maps and their impact on kernel geometry, expressivity, and classification performance. The experiments investigate quantum kernel methods based on the Loschmidt Echo formulation, implemented using PennyLane, and evaluate multiple quantum embeddings across different datasets and input dimensions. For each configuration (dataset × embedding × dimension), the dataset includes: - quantum kernel matrices,- classification results (SVM with precomputed kernels),- kernel target alignment (KTA),- effective dimension of the kernel,- model complexity measures,- geometric difference metrics,- aggregated statistical summaries,- configuration snapshots ensuring full reproducibility. The dataset is structured to support full reproducibility of the experimental pipeline and enables independent analysis of quantum kernel geometry and learning performance. All results correspond to benchmark experiments conducted on datasets from the PennyLane QML benchmark collection. This dataset accompanies a study on the relationship between quantum embeddings, kernel geometry, and machine learning performance.

•Jun 19, 2026•0 citations

Empirical Evaluation of Quantum Kernel Methods for Classification: A Comparative Study with Classical Kernels

Ronak Bhalgami

Quantum kernel methods are the most concrete near-term proposal for quantum machine learning on classical data, yet published benchmarks disagree on whether they outperform tuned classical kernels, and the disagreement cannot be resolved from the literature because the comparisons are not protocol-controlled. We argue that the obstacle is one of measurement, not theory: the usual setup fixes the quantum feature map's data scaling while grid-searching the classical kernel's hyperparameters, recomputes nothing per fold, and reports cost in simulator time. We therefore treat the comparison itself as the object of study and impose three controls. The bandwidth is tuned on the same grid budget as the classical $γ$, the kernel is recomputed inside every fold from fold-fitted preprocessing, and cost is counted in hardware circuit evaluations. On top of these we run a factorial design of 324 cells in which dimensionality, training size, class separability and data structure are explicit covariates, together with four standard datasets, a shots-versus-gate-noise ablation, and a cost sweep. The dominant effect we measure is not quantum: tuning the feature map's bandwidth is worth 0.120 accuracy on synthetic data and 0.339 on real data, exceeding every quantum-classical difference we observe. Tuned quantum kernels are then statistically indistinguishable from a tuned RBF on all four real datasets. The advantage on synthetic data is confined almost entirely to data generated by the feature map itself, and it shrinks with dimensionality (-0.0045 per qubit) while growing with sample size. Finite sampling costs an order of magnitude more accuracy than gate and readout noise, and a fidelity kernel needs $5\times10^{7}$ circuits at n = 10,000. Protocol variance is thus sufficient to explain the field's contradictory conclusions.

quant-pharXiv:2610.42029•Nov 2, 2026•0 citations

The Inductive Bias of Quantum Kernels

Jonas M. Kübler, Simon Buchholz, Bernhard Schölkopf

It has been hypothesized that quantum computers may lend themselves well to applications in machine learning. In the present work, we analyze function classes defined via quantum kernels. Quantum computers offer the possibility to efficiently compute inner products of exponentially large density operators that are classically hard to compute. However, having an exponentially large feature space renders the problem of generalization hard. Furthermore, being able to evaluate inner products in high dimensional spaces efficiently by itself does not guarantee a quantum advantage, as already classically tractable kernels can correspond to high- or infinite-dimensional reproducing kernel Hilbert spaces (RKHS). We analyze the spectral properties of quantum kernels and find that we can expect an advantage if their RKHS is low dimensional and contains functions that are hard to compute classically. If the target function is known to lie in this class, this implies a quantum advantage, as the quantum computer can encode this inductive bias, whereas there is no classically efficient way to constrain the function class in the same way. However, we show that finding suitable quantum kernels is not easy because the kernel evaluation might require exponentially many measurements. In conclusion, our message is a somewhat sobering one: we conjecture that quantum machine learning models can offer speed-ups only if we manage to encode knowledge about the problem at hand into quantum circuits, while encoding the same bias into a classical model would be hard. These situations may plausibly occur when learning on data generated by a quantum process, however, they appear to be harder to come by for classical datasets.

arXiv:2106.03747•Jun 7, 2021•100 citations

Quantum Machine Learning in Feature Hilbert Spaces

Maria Schuld, Nathan Killoran

A basic idea of quantum computing is surprisingly similar to that of kernel methods in machine learning, namely, to efficiently perform computations in an intractably large Hilbert space. In this Letter we explore some theoretical foundations of this link and show how it opens up a new avenue for the design of quantum machine learning algorithms. We interpret the process of encoding inputs in a quantum state as a nonlinear feature map that maps data to quantum Hilbert space. A quantum computer can now analyze the input data in this feature space. Based on this link, we discuss two approaches for building a quantum model for classification. In the first approach, the quantum device estimates inner products of quantum states to compute a classically intractable kernel. The kernel can be fed into any classical kernel method such as a support vector machine. In the second approach, we use a variational quantum circuit as a linear model that classifies data explicitly in Hilbert space. We illustrate these ideas with a feature map based on squeezing in a continuous-variable system, and visualize the working principle with two-dimensional minibenchmark datasets.

arXiv:1803.07128•Physical Review Letters•Feb 1, 2019•2,212 citations

Investigating Quantum Kernel Feature Maps and Kernel Alignment for Multivariate Spatio-Temporal Pattern Classification

Vishwambhar Pathak, Nisheeth Joshi, Prabhat Mahanti

AI-driven solutions and simulations often involve spatio-temporal multivariate signals. However, such real-life signals are mostly nonlinear, noisy, and high-dimensional. Fetching sparse and subtle spatio-temporal patterns from them for descriptive and predictive applications poses challenge to traditional machine learning models. With recent practicable advancements in quantum computing technologies, quantum kernel methods have been exploited to map classical data into high-dimensional Hilbert spaces which transform complex relationships into separable patterns. In this work we investigate the effectiveness of quantum kernel methods for the case of EEG signal classification and compares their performance with classical radial basis function (RBF) kernels. Quantum encoding was performed using angle-encoding and necessary circuit elements including swap-test. The investigation was conducted using two sets of synthetic EEG datasets: (1) simple sinusoidal signals and (2) nonlinear signals with phase and amplitude variability. Principal Component Analysis was used to reduce dimensionality before kernel construction. For contrasting the Classical vs Quantum workflows, feature maps produced by Classical and Quantum kernel matrices were computed and fed to Support Vector Machine (SVM) classifier. The workflow then evaluated Kernel-target alignment for estimation of the correspondence between kernel similarity structures and class labels. For simple sinusoidal signals, training with the classical kernels achieved accuracies up to 0.967 for 8 principal component features as compared to 0.86 accuracy with the quantum kernels. However, for nonlinear EEG signals, while both workflows gave lower accuracies, still quantum kernels yielded slightly better accuracy 0.46 as compared to 0.40 by the classical kernels. The kernel-target alignment was mostly better with classical kernel, with slightly better for quantum kernel in one case. The slight improvement with quantum kernel in non-linear signal case suggests that quantum kernels may provide advantages in modelling complex nonlinear relationships. Limited alignment and limited accuracy improvement with the quantum kernel indicate need for further improvement in quantum-feature-map design to achieve quantum advantage.

•May 1, 2026•0 citations

Quantum Feature Spaces and Kernels

Hamid D. Ismail

This chapter develops the theory and practice of quantum feature maps and quantum kernels as a central framework in quantum machine learning (QML). Building on the data-encoding methods introduced in the previous chapter, it explains how classical inputs are embedded into quantum states and how similarity between data points is then defined through overlaps in Hilbert space. This chapter introduces the quantum kernel as the squared overlap between encoded states, establishes its fundamental algebraic properties, and shows how it satisfies the conditions required for direct use in classical kernel-learning methods such as support vector machines, kernel ridge regression, and Gaussian process models. A major focus is placed on the circuit-level realization and estimation of quantum kernels. This chapter presents kernel evaluation through encoding unitaries and examines three principal estimation methods: the SWAP test, the Hadamard test, and the direct return-probability method, with attention to their trade-offs in qubit overhead, circuit depth, and suitability for NISQ hardware. It also develops the geometric interpretation of quantum kernels in Hilbert space, relating kernel values to angular separation, decision boundaries, expressivity, and generalization. Together, these ideas provide a unified foundation for understanding how quantum circuits, feature maps, and classical learning algorithms interact in hybrid quantum-classical kernel models.

CRC Press eBooks•Nov 4, 2026•0 citations
Quantum kernels are among the most promising methods in quantum machine learning, providing a natural way to embed data into high-dimensional Hilbert spaces via quantum circuits and capturing non-linear relationships in data. Their learning behaviour is governed by the function class they represent, which is formally characterized by the structure of the associated Reproducing Kernel Hilbert Space (RKHS). However, a theoretical characterization of this function class, and of how it depends on the choice of feature map and its hyperparameters, is still lacking. In this work, we focus on the Instantaneous Quantum Polynomial (IQP) feature map and derive a Fourier representation of the associated RKHS generators as finite sums of trigonometric functions evaluated on polynomially transformed input variables, with a finite set of accessible frequencies determined by the circuit construction and scaled by the bandwidth hyperparameter. This representation identifies three bandwidth regimes with qualitatively distinct expressive properties from localised, through multimodal, to periodic. Numerical experiments on synthetic datasets with controlled periodic and multimodal structure, across data dimensions and feature-map configurations, reproduce the three regimes in both supervised classification and unsupervised anomaly detection, benchmarked against a classical Gaussian kernel. These results make the relationship between expressivity and data structure explicit for IQP kernels: expressive power translates into a practical benefit only when the accessible frequencies match with the structure of the data, which provides a criterion for selecting the bandwidth when the periodic or multimodal structure of data is known.
The paper asks how the function class represented by an IQP fidelity kernel depends on its feature map and hyperparameters, particularly input bandwidth bbb. It studies this through the generators k(x,⋅)k(x,\cdot)k(x,⋅) of the associated reproducing kernel Hilbert space (RKHS).
Extending an existing circuit Fourier-expansion framework, it represents the generators as finite trigonometric sums evaluated on polynomial transformations of the inputs of degree at most two. For Pauli-ZZZ encodings, elementary frequency differences lie in {−2,0,2}\{-2,0,2\}{−2,0,2}; repeated encodings permit finite sums of these differences. The representation motivates three qualitative bandwidth regimes: Gaussian-like/local at small bbb, multimodal at intermediate bbb, and strongly oscillatory at large bbb. Polynomial coordinate transformations mean the kernels are generally not exactly periodic in the original inputs.
Numerical tests use synthetic periodic and Gaussian-mixture datasets, SVM classification and one-class SVM anomaly detection, with D=2,4,6D=2,4,6D=2,4,6 qubits, one to three circuit repetitions, and linear or full entanglement. On periodic classification data, a representative two-qubit configuration achieves F1 scores of 0.750.750.75–0.840.840.84 when bbb matches the label frequency, versus 0.460.460.46–0.720.720.72 otherwise. On multimodal data, performance typically peaks around b=1b=1b=1–1.51.51.5; quantum and Gaussian-kernel performance ranges largely overlap.
The stated limitations are that numerical evidence is restricted to controlled synthetic datasets and the theoretical characterization concerns IQP feature maps. Applicability to real-world data and other feature-map families remains open.
This dataset contains the complete experimental results and supporting data generated in the study of quantum feature maps and their impact on kernel geometry, expressivity, and classification performance. The experiments investigate quantum kernel methods based on the Loschmidt Echo formulation, implemented using PennyLane, and evaluate multiple quantum embeddings across different datasets and input dimensions. For each configuration (dataset × embedding × dimension), the dataset includes: - quantum kernel matrices,- classification results (SVM with precomputed kernels),- kernel target alignment (KTA),- effective dimension of the kernel,- model complexity measures,- geometric difference metrics,- aggregated statistical summaries,- configuration snapshots ensuring full reproducibility. The dataset is structured to support full reproducibility of the experimental pipeline and enables independent analysis of quantum kernel geometry and learning performance. All results correspond to benchmark experiments conducted on datasets from the PennyLane QML benchmark collection. This dataset accompanies a study on the relationship between quantum embeddings, kernel geometry, and machine learning performance.
Quantum kernel methods are the most concrete near-term proposal for quantum machine learning on classical data, yet published benchmarks disagree on whether they outperform tuned classical kernels, and the disagreement cannot be resolved from the literature because the comparisons are not protocol-controlled. We argue that the obstacle is one of measurement, not theory: the usual setup fixes the quantum feature map's data scaling while grid-searching the classical kernel's hyperparameters, recomputes nothing per fold, and reports cost in simulator time. We therefore treat the comparison itself as the object of study and impose three controls. The bandwidth is tuned on the same grid budget as the classical γγγ, the kernel is recomputed inside every fold from fold-fitted preprocessing, and cost is counted in hardware circuit evaluations. On top of these we run a factorial design of 324 cells in which dimensionality, training size, class separability and data structure are explicit covariates, together with four standard datasets, a shots-versus-gate-noise ablation, and a cost sweep. The dominant effect we measure is not quantum: tuning the feature map's bandwidth is worth 0.120 accuracy on synthetic data and 0.339 on real data, exceeding every quantum-classical difference we observe. Tuned quantum kernels are then statistically indistinguishable from a tuned RBF on all four real datasets. The advantage on synthetic data is confined almost entirely to data generated by the feature map itself, and it shrinks with dimensionality (-0.0045 per qubit) while growing with sample size. Finite sampling costs an order of magnitude more accuracy than gate and readout noise, and a fidelity kernel needs 5×1075\times10^{7}5×107 circuits at n = 10,000. Protocol variance is thus sufficient to explain the field's contradictory conclusions.
It has been hypothesized that quantum computers may lend themselves well to applications in machine learning. In the present work, we analyze function classes defined via quantum kernels. Quantum computers offer the possibility to efficiently compute inner products of exponentially large density operators that are classically hard to compute. However, having an exponentially large feature space renders the problem of generalization hard. Furthermore, being able to evaluate inner products in high dimensional spaces efficiently by itself does not guarantee a quantum advantage, as already classically tractable kernels can correspond to high- or infinite-dimensional reproducing kernel Hilbert spaces (RKHS). We analyze the spectral properties of quantum kernels and find that we can expect an advantage if their RKHS is low dimensional and contains functions that are hard to compute classically. If the target function is known to lie in this class, this implies a quantum advantage, as the quantum computer can encode this inductive bias, whereas there is no classically efficient way to constrain the function class in the same way. However, we show that finding suitable quantum kernels is not easy because the kernel evaluation might require exponentially many measurements. In conclusion, our message is a somewhat sobering one: we conjecture that quantum machine learning models can offer speed-ups only if we manage to encode knowledge about the problem at hand into quantum circuits, while encoding the same bias into a classical model would be hard. These situations may plausibly occur when learning on data generated by a quantum process, however, they appear to be harder to come by for classical datasets.
A basic idea of quantum computing is surprisingly similar to that of kernel methods in machine learning, namely, to efficiently perform computations in an intractably large Hilbert space. In this Letter we explore some theoretical foundations of this link and show how it opens up a new avenue for the design of quantum machine learning algorithms. We interpret the process of encoding inputs in a quantum state as a nonlinear feature map that maps data to quantum Hilbert space. A quantum computer can now analyze the input data in this feature space. Based on this link, we discuss two approaches for building a quantum model for classification. In the first approach, the quantum device estimates inner products of quantum states to compute a classically intractable kernel. The kernel can be fed into any classical kernel method such as a support vector machine. In the second approach, we use a variational quantum circuit as a linear model that classifies data explicitly in Hilbert space. We illustrate these ideas with a feature map based on squeezing in a continuous-variable system, and visualize the working principle with two-dimensional minibenchmark datasets.
AI-driven solutions and simulations often involve spatio-temporal multivariate signals. However, such real-life signals are mostly nonlinear, noisy, and high-dimensional. Fetching sparse and subtle spatio-temporal patterns from them for descriptive and predictive applications poses challenge to traditional machine learning models. With recent practicable advancements in quantum computing technologies, quantum kernel methods have been exploited to map classical data into high-dimensional Hilbert spaces which transform complex relationships into separable patterns. In this work we investigate the effectiveness of quantum kernel methods for the case of EEG signal classification and compares their performance with classical radial basis function (RBF) kernels. Quantum encoding was performed using angle-encoding and necessary circuit elements including swap-test. The investigation was conducted using two sets of synthetic EEG datasets: (1) simple sinusoidal signals and (2) nonlinear signals with phase and amplitude variability. Principal Component Analysis was used to reduce dimensionality before kernel construction. For contrasting the Classical vs Quantum workflows, feature maps produced by Classical and Quantum kernel matrices were computed and fed to Support Vector Machine (SVM) classifier. The workflow then evaluated Kernel-target alignment for estimation of the correspondence between kernel similarity structures and class labels. For simple sinusoidal signals, training with the classical kernels achieved accuracies up to 0.967 for 8 principal component features as compared to 0.86 accuracy with the quantum kernels. However, for nonlinear EEG signals, while both workflows gave lower accuracies, still quantum kernels yielded slightly better accuracy 0.46 as compared to 0.40 by the classical kernels. The kernel-target alignment was mostly better with classical kernel, with slightly better for quantum kernel in one case. The slight improvement with quantum kernel in non-linear signal case suggests that quantum kernels may provide advantages in modelling complex nonlinear relationships. Limited alignment and limited accuracy improvement with the quantum kernel indicate need for further improvement in quantum-feature-map design to achieve quantum advantage.
This chapter develops the theory and practice of quantum feature maps and quantum kernels as a central framework in quantum machine learning (QML). Building on the data-encoding methods introduced in the previous chapter, it explains how classical inputs are embedded into quantum states and how similarity between data points is then defined through overlaps in Hilbert space. This chapter introduces the quantum kernel as the squared overlap between encoded states, establishes its fundamental algebraic properties, and shows how it satisfies the conditions required for direct use in classical kernel-learning methods such as support vector machines, kernel ridge regression, and Gaussian process models. A major focus is placed on the circuit-level realization and estimation of quantum kernels. This chapter presents kernel evaluation through encoding unitaries and examines three principal estimation methods: the SWAP test, the Hadamard test, and the direct return-probability method, with attention to their trade-offs in qubit overhead, circuit depth, and suitability for NISQ hardware. It also develops the geometric interpretation of quantum kernels in Hilbert space, relating kernel values to angular separation, decision boundaries, expressivity, and generalization. Together, these ideas provide a unified foundation for understanding how quantum circuits, feature maps, and classical learning algorithms interact in hybrid quantum-classical kernel models.