Representation Geometry and Inductive Biases in Quantum Classifiers across Classical Visual Benchmarks
Abstract
Quantum Machine Learning (QML) has a theoretical advantage over classical algorithms for specific synthetic data that is computationally hard. However, it remains unclear whether such theoretical advantages play a meaningful role for realistic classical data, such as natural images. Quantum structural properties induced by unitary operations or entanglement introduce inductive biases that differ from those of classical models, potentially leading to distinct learning behavior on real-world data. In a fully quantum learning pipeline, classical real-world data are first encoded into quantum states via a quantum feature mapping, analogous to a classical feature extractor, and a quantum classifier then classifies the resulting states. Quantum feature mappings generate representations that are difficult to realize classically, while also inducing representation-level biases that do not necessarily match the structure of realistic data. To enable a fair comparison at the classifier level, we focus on the classifier and use classical feature extractors that already provide powerful representations for natural images. By controlling the data scale and supervision strength, we analyze how the unitary structure of quantum classifiers affects generalization, and show that their effectiveness depends strongly on the geometry of the input representations. These results indicate that quantum classifiers benefit from classical features that are aligned with quantum feature spaces. Our analysis shows that kernel alignment identifies representations compatible with quantum feature spaces, providing insights into translating classical models into quantum algorithms.
Related
- A comparative study of variational quantum classifiers and classical machine learning for tW−tt̅ discrimination at the LHC
- A Dynamic Intermediate Representation for Hybrid Quantum-Classical Programs
- Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics
- Qaiji IR: An Eight-Layer Intermediate Representation Family for Hybrid Quantum-Classical Compilation
- Warp-Geo: Differentiable Geometry Representation for Dynamic-Boundary Simulation and Shape Optimization
BibTeX
@misc{yamaoka2026representation,
title = {{Representation Geometry and Inductive Biases in Quantum Classifiers across Classical Visual Benchmarks}},
author = {Yu Yamaoka and Weng Ian Chan and Kosuke Mitarai},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.17722},
eprint = {2610.17722},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.17722},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Representation Geometry and Inductive Biases in Quantum Classifiers across Classical Visual Benchmarks
AU - Yu Yamaoka
AU - Weng Ian Chan
AU - Kosuke Mitarai
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2610.17722
UR - https://arxiv.org/abs/2610.17722
AB - Quantum Machine Learning (QML) has a theoretical advantage over classical algorithms for specific synthetic data that is computationally hard. However, it remains unclear whether such theoretical advantages play a meaningful role for realistic classical data, such as natural images. Quantum structural properties induced by unitary operations or entanglement introduce inductive biases that differ from those of classical models, potentially leading to distinct learning behavior on real-world data. In a fully quantum learning pipeline, classical real-world data are first encoded into quantum states via a quantum feature mapping, analogous to a classical feature extractor, and a quantum classifier then classifies the resulting states. Quantum feature mappings generate representations that are difficult to realize classically, while also inducing representation-level biases that do not necessarily match the structure of realistic data. To enable a fair comparison at the classifier level, we focus on the classifier and use classical feature extractors that already provide powerful representations for natural images. By controlling the data scale and supervision strength, we analyze how the unitary structure of quantum classifiers affects generalization, and show that their effectiveness depends strongly on the geometry of the input representations. These results indicate that quantum classifiers benefit from classical features that are aligned with quantum feature spaces. Our analysis shows that kernel alignment identifies representations compatible with quantum feature spaces, providing insights into translating classical models into quantum algorithms.
ER -