Machine learning (ML) techniques provide a powerful alternative for solving scattering problems by leveraging data-driven approaches to model complex interactions. Unlike traditional methods, ML models learn patterns and relationships directly from data, enabling rapid predictions without explicit physics-based formulations. However, integrating physics-based principles into ML models, as demonstrated in [1-3], such as incorporating constraints from Maxwell’s equations, enhances interpretability and ensures physically consistent solutions. For instance, the study in [4] introduces a new method for electromagnetic inverse scattering problem which gains advantages by deep learning as well as the paradigms of ‘virtual experiments’. These methods excel in handling large-scale problems, noisy data, and nonlinear scattering scenarios.
[1] S. Zumbo, S. Mandija, F. Meliadó, C. A. T. van den Berg, T. Isernia and M. T. Bevacqua, “Application of Supervised Descent Method to MRI Electrical Properties Tomography,” 2022 16th European Conference on Antennas and Propagation (EuCAP), Madrid, Spain, 2022, pp. 1-5, doi: 10.23919/EuCAP53622.2022.9768930. [click here]
[2] S. Zumbo, S. Mandija, E.F. Meliado, P. Stijnman , T.G. Meerbothe, C.A.T. van den Berg, T. Isernia, M. T. Bevacqua, “Unrolled Optimization via Physics-Assisted Convolutional Neural Network for MR-Based Electrical Properties Tomography: A Numerical Investigation,” IEEE Open J Eng Med Biol. 2024 May 20;5:505-513. doi: 10.1109/OJEMB.2024.3402998. [click here]
[3] S. Zumbo, S. Mandija, T. Isernia, & M.T. Bevacqua, “MiPhDUO: microwave imaging via physics-informed deep unrolled optimization,”2024 Inverse Problems, 40(4), 045017. [click here]
[4] M. T. Bevacqua, C. Ieracitano, N. Mammone, F. C. Morabito, T. Isernia and L. Di Donato, “Electromagnetic Inverse Scattering via Deep Learning Enhanced by Virtual Experiments,” 2023 Photonics & Electromagnetics Research Symposium (PIERS), Prague, Czech Republic, 2023, pp. 1231-1236, doi: 10.1109/PIERS59004.2023.10221556. [click here]