Towards Quantum Operator-Valued Kernels

Hachem Kadri, Joachim Tomasi, Yuka Hashimoto, Sandrine AnthoinearXiv 2025 · DOI · arXiv

Introduction

Quantum Machine Learning (QML) has reached a pivotal moment. While initial enthusiasm focused on potential speed-ups for classical tasks, recent findings suggest that standard quantum kernels—specifically scalar-valued ones—often fail to outperform classical counterparts on simple classification problems. This position paper argues for a paradigm shift: moving beyond scalar similarities to Quantum Operator-Valued Kernels (QOVKs). By leveraging quantum entanglement to link input and output spaces, QOVKs are designed to address "structured prediction" tasks where classical methods and standard quantum kernels struggle.

The conceptual evolution from standard QML to Quantum Operator-Valued Kernels.

The Expressivity Gap in Scalar Kernels

Most current QML research relies on scalar-valued kernels, which output a single real number representing the similarity between two data points. While effective for binary classification, these kernels are inherently limited when dealing with high-dimensional, complex outputs. Tasks such as multitask learning or structured prediction—where the goal is to predict objects like graphs, curves, or quantum states—require a more expressive framework. QOVKs fill this gap by outputting operators (matrices) that can capture intricate dependencies between multiple output variables simultaneously.

Formalizing Quantum Operator-Valued Kernels

The authors introduce a framework for QOVKs that extends the traditional kernel method into the operator domain.

  1. Operator-Valued Kernels (OVKs): Unlike a standard kernel $k(x, z) \in \mathbb{R}$, an OVK returns an operator $K(x, z)$ acting on an output space. This allows the model to learn correlations across the entire output structure.
  2. Entangled QOVKs: The "quantum secret sauce" lies in non-separable (entangled) unitary transformations. By entangling the input data encoding with the output space, the model uses entanglement as a computational resource to represent correlations that are classically intractable.
  3. Implementation: The paper proposes a generalized "swap test" circuit to prepare quantum states corresponding to these operator values, facilitating integration into hybrid quantum-classical workflows.

The Expressivity Hierarchy

A central contribution of this work is the establishment of a clear hierarchy of quantum kernel expressivity. As shown in the figure below, QOVKs represent the most flexible class of kernels, encompassing and extending beyond standard scalar and separable quantum kernels.

The expressivity hierarchy of quantum kernels, placing QOVKs at the top.

Experimental Validation: Quantum Channel Estimation

To demonstrate the power of QOVKs, the authors tackled the task of Quantum Channel Estimation—predicting the behavior of a noisy Pauli channel. Using a dataset of random density matrices, they compared Entangled QOVKs against standard separable approaches.

  • Quantitative Performance: The Entangled QOVK achieved a recovery error (Frobenius norm) of 0.15 ± 0.05, a significant improvement over the 0.55 ± 0.25 error seen with scalar-valued methods.
  • Structural Accuracy: Qualitative analysis revealed that the Entangled QOVK successfully recovered the underlying "block" structure of the Pauli channel, a feat the scalar-valued kernel failed to achieve.

Discussion and Future Outlook

While QOVKs offer a promising path toward quantum advantage in structured prediction, several challenges remain:

  • Computational Complexity: The resulting kernel matrices are significantly larger ($np \times np$), posing a classical bottleneck. Future integration with quantum linear system solvers (like HHL) may be necessary.
  • The Pre-image Problem: Mapping kernel results back to physical objects (e.g., specific molecules) remains non-trivial on quantum hardware.
  • Generalization: Like all high-dimensional models, QOVKs must be carefully designed to avoid "exponential concentration," where the kernel matrix loses its discriminative power as qubit counts increase.

By shifting the focus from simple classification speed to the complexity of structured tasks, QOVKs open a new frontier for meaningful quantum advantage in machine learning.

Summary updated

Abstract

Quantum kernel functions built using quantum-mechanical principles and have emerged as a centerpiece of quantum machine learning. The initial enthusiasm for quantum kernel machines has been tempered by recent studies suggesting that quantum kernels could not offer significant computational or statistical advantages when learning from classical data. However, most of the research in this area has been devoted to scalar-valued kernels in standard classification or regression settings for which classical kernel methods are efficient and effective, leaving very little room for improvement with quantum kernels. In this position paper, we argue that progress in this field requires moving beyond scalar-valued kernels toward more expressive kernel frameworks. Scalar-valued kernels lack the degrees of freedom necessary to fully exploit intrinsically quantum resources such as entanglement and are not rich enough to deal with complex learning tasks where classical learning methods struggle. Building on recent advances in operator-valued kernel learning and $C^*$-algebraic kernel representations, we propose a roadmap for designing quantum kernels capable of leveraging entanglement and non-commutative structures to tackle complex structured prediction problems. To support this viewpoint, we present an initial proof-of-concept illustrating how quantum operator-valued kernel formulations can reveal structural dependencies that remain difficult to access for scalar-valued kernel methods. This shift in focus could open a pathway toward a new generation of quantum kernel machines and a more faithful exploration of their potential advantages.
BibTeX
@misc{kadri2025towards,
  author = {Hachem Kadri and Joachim Tomasi and Yuka Hashimoto and Sandrine Anthoine},
  title = {{Towards Quantum Operator-Valued Kernels}},
  year = {2025},
  eprint = {2506.03779},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2506.03779},
}
Figure