Broadband Field Prediction via Parameter-Efficient Electromagnetic Surrogate Solver

초록

We propose a parameter-efficient electromagnetic surrogate solver that predicts broadband field distributions from discrete wavelength data. Using wave-informed multiplicative encoding and Fourier group convolutional shuffling, our approach reduces parameters by 74% and improves the accuracy of untrained wavelength prediction by 80.5%, outperforming state-of-the-art models.