Adaptive Domain Decomposition PINNs Handle Sparse Sensor Data for Traffic State Estimation
arXiv·low signal
Addresses the failure mode of physics-informed neural networks that over-smooth traffic dynamics when sensor data is sparse, by adaptively decomposing the spatiotemporal domain into subregions. Achieves accurate traffic state estimation from fixed-sensor deployments that would defeat standard PINNs. Niche application but demonstrates a generalizable pattern for making PINNs robust under data sparsity.