Speaker: Prof. Xudong CHEN, National University of Singapore

Inviter: 张波研究员

Title: Physics-Assisted Machine Learning for Solving Forward and Inverse Scattering Problems

Language: Chinese

Time & Venue: 2025.05.28 10:00-11:00 南楼620

Abstract: In recent years, machine learning (ML) has garnered significant interest for solving forward and inverse scattering problems. However, many approaches treat ML as a black box, bypassing decades of wave‐physics insights, mathematical structure, and computational techniques. This talk demonstrates how embedding domain knowledge—and leveraging mathematical, physical, and engineering intuition—within machine learning yields more efficient, elegant solutions. Forward problem: Inspired by the volume integral equation discretization via the method of moments, we introduce an Electric Flux Density Learning Method (EFDLM), a cascaded neural network architecture, that accurately computes 3D electromagnetic scattering from lossless dielectric objects. Inverse problem: By fully understanding the forward model, we design physics‐guided inputs and outputs for our neural network, showing that this guided approach dramatically enhances reconstruction resolution. We conclude by distilling the key lessons learned from tackling both forward and inverse scattering challenges.