Speaker
Description
Quantum kernel methods rely on data-dependent feature maps to embed classical inputs into quantum states, where the induced fidelity kernel defines the geometry used by a classical learning algorithm. The choice of feature map is therefore a central bottleneck in quantum machine learning: generic encodings are hardware-friendly but may impose weak inductive bias, while structure-aware encodings can be effective but typically require prior knowledge of the data. In this work, we formulate quantum feature-map design as a supervised, hardware-constrained architecture search problem. We propose Quantum Feature Map Search (QFMS), an automated framework for discovering dataset-adapted quantum encodings under explicit restrictions on qubit count, circuit depth, native gates, and device connectivity. Candidate feature maps are represented as discrete chromosomes encoding gate placements, fixed rotations, feature-dependent rotations, entangling operations, and a global bandwidth parameter. Each chromosome is decoded into a hardware-compliant circuit, evaluated through the fidelity kernel it induces on the training data, and scored using an SVM-derived loss. A genetic algorithm then searches directly over encoding structures, optimizing the task-dependent kernel geometry rather than an abstract circuit expressivity measure. We evaluate the methodology through a staged protocol involving a group-structured benchmark inspired by covariant quantum kernels, the high-dimensional Madelon synthetic dataset, and the HIGGS real-world classification task from high-energy physics. We also introduce a transfer-expansion strategy that embeds feature maps found under smaller budgets into larger circuit families. The goal is not to claim quantum advantage, but to provide a systematic route toward dataset-adapted, hardware-aware quantum kernels comparable against classical baselines.
| Topic | Quantum kernel methods and their applications |
|---|---|
| Additional Topic | Hybrid quantum-classical approaches for learning and optimization |