arXiv:2610.24403
Prototype-Guided Optimization of Quantum Data Embeddings with Linear Cost
Seokhoon Jeong, Daniel K. Park
Abstract
In supervised quantum machine learning with classical data, classification performance critically depends on the geometry induced by quantum data embedding. Existing data-adaptive approaches rely on pairwise fidelity evaluations between samples, which scale as $O(B^2)$ for $B$ samples and create a computational bottleneck in iterative embedding optimization. To address this issue, we propose a prototype-guided framework for efficient data-adaptive quantum embedding. Using class-prototype separation and sample-to-prototype compactness, we derive a measurable lower certificate for class-ensemble distinguishability and construct a prototype-based objective requiring only $O(B)$ fidelity evaluations. We integrate this objective into an end-to-end pipeline that combines task-specific quantum circuit architecture optimization with continuous refinement of the data-embedding parameters. Across five benchmark datasets with up to 20 qubits, our method retains competitive quantum kernel support vector machine (SVM) performance while reducing fidelity-evaluation cost, with particularly strong gains over budget-matched pairwise optimization on T-MNIST. Geometric analyses examine how the learned embeddings relate to kernel alignment and classification margins, revealing dataset-dependent benefits and limitations. We further derive a prototype-based classifier in the reproducing kernel Hilbert space that provides an explicit witness for the SVM objective and a low-cost proxy for model selection, and extend the framework to multiple prototypes for multimodal class structures.
AI Summary
The paper addresses the fidelity-evaluation cost of supervised quantum embedding optimization: evaluating all sample pairs costs $O(B^2)$ per objective evaluation. It replaces pairwise comparisons with sample-to-class-prototype compactness and prototype separation, requiring $B+\binom{K_{\mathrm{tot}}}{2}$ fidelities, or $O(B)$ for a fixed number of prototypes. For binary pure-state embeddings, it proves the certificate $\|p_+\bar\rho_+-p_-\bar\rho_-\|_1\geq\sqrt{1-4p_+p_-F_c}-2\sum_y p_y\bar r_y^{\mathrm{tr}}$ using the reverse triangle inequality and trace-norm convexity. A multi-prototype extension replaces prototype states with weighted mixtures. These bounds motivate fidelity-based surrogate losses. A prototype kernel classifier also supplies an explicit feasible witness bounding the optimal soft-margin SVM objective from above, and a linear-cost model-selection proxy. Experiments combine genetic circuit search with neural input-map refinement on five datasets using 4–20 qubits and noiseless statevector simulation. Across ten seeds, the largest improvement over budget-matched sampled-pair optimization occurs on five-class T-MNIST: test MCC increases from $0.6357$ to $0.7606$ after circuit search and from $0.6423$ to $0.7787$ after refinement. Other datasets show mixed differences. On GLASS, multiple prototypes increase MCC from approximately $0.592$ to $0.647$. The certificate is nonpositive on three datasets, and better average geometry or kernel alignment does not consistently improve SVM performance; difficult boundary samples can deteriorate. The objective is a surrogate, benefits depend on prototype quality and class structure, and the multi-prototype experiment uses known subclass labels. Automatic prototype construction, adaptive assignments, and finite-shot noisy-hardware evaluation remain open.
In this work, we used generative AI tools to create or edit software code and make up draft parts of a research paper.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 26 pages, 5 figures, 5 tables