Kernel-based Dequantization of Variational Quantum Machine Learning

Explore kernel-based dequantization methods for variational quantum machine learning in this 17-minute conference presentation from QTML 2025. Learn how classical kernel-based algorithms can match the performance of quantum models based on parameterized quantum circuits (PQCs), including quantum neural networks and quantum kernel methods. Discover the development of trigonometric kernel families that capture function classes expressible by quantum models and understand how Random Fourier Features (RFF) can approximate these kernels. Examine theoretical bounds on risk differences between RFF-approximated classical models and their quantum counterparts in regression and classification tasks, and identify sufficient conditions for successful dequantization. Investigate cases where exact kernels can be computed efficiently through tensor networks, eliminating approximation requirements. Gain insights into designing PQC-based QML models that avoid dequantization and understand how proposed dequantization schemes serve as independent heuristic classical algorithms before deploying resource-intensive quantum machine learning models.