Quantum Spectral Model: Data Reuploading with Input-Conditioned Frequency Support

The paper introduces Quantum Spectral Models (QSMs), which construct the generator of a quantum data-encoding unitary directly from each input matrix. It studies symmetric, global block, and non-overlapping patch-local Hamiltonian variants. Their outputs have truncated Fourier representations whose candidate phase carriers are input-dependent spectral gaps, while spectral subspaces help determine coefficients. Across two matrix representations of Pendigits and two controlled synthetic tasks based on spectral statistics, QSM variants achieved the best mean test accuracy among the evaluated quantum models at the largest circuit depth. Patch-local QSM led on Pendigits, while global block QSM led on the synthetic spectral tasks.