Speaker
Description
The classification performance of quantum supervised learning on classical data is fundamentally controlled by the feature map that embeds input into quantum states. To promote quantum state distinguishability, we propose an energy-based generative learning framework that synthesizes gate sequences to optimize embedding structures and refine data-tailored parameters using a fidelity-based surrogate objective. Moreover, we explain the saturation of classification performance by extending quantum risk bounds into the classical input space, demonstrating that the achievable empirical risk across an embedding family is bounded by the 1-Wasserstein distance. Our research establishes a data-driven quantum embedding framework while showing that classical data geometry provides an a-priori diagnostic for optimization gains.
| Topic | Hybrid quantum-classical approaches for learning and optimization |
|---|---|
| Additional Topic | Quantum kernel methods and their applications |