arxivcs.LG2026-07-06
Target-Guided Selective Reweighting for Physics-Informed Neural Network Inverse Problems: A Transfer Learning Approach
Qian Hu, Bin Fan, Yao Xiao, Zhicheng Lin, Meixin Xiong
Physics-informed neural networks (PINNs) encounter ill-posed optimization, loss competition, and parameter compensation in partial differential equation (PDE) inverse problems. Transfer learning can reuse representations from source tasks, but direct fine-tuning may introduce neg…