{"id":"arxiv/2606.06171","url":"https://arxiv.org/abs/2606.06171","tags":["stat.ML","cs.LG","math.NA","physics.comp-ph"],"score":0,"title":"Effective Dimensionality as an Operator Invariant for Physics-Preserving Constraint Adaptation in Physics-Informed Neural Networks","domain":"stat.ML","source":"arxiv","authors":["Cornelius Otchere","Michael Shields"],"reviews":0,"subject":"stat.ML","summary":"Physics-Informed Neural Networks inherently suffer from task interference because they rely on a shared parameter space to satisfy both governing differential equations and boundary conditions. We analyze this structural conflict using the Fisher Information Matrix to quantify the effective degrees of freedom ($d_{eff}$) in a physics-constrained model. Unlike the classical $d_{eff}$ which measures how many parameter directions are informed by data against a statistical prior, our $d_{eff}$ measures the dimension of the parameter directions unconstrained by the differential operator. For…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-06-04","reactions":{}}