Supervised initialization of tension parameters in non-uniform non-stationary subdivision schemes

Abstract

In this work, we propose a supervised learning approach for initialising local tension parameters in non-uniform and non-stationary subdivision schemes. A neural network generates an initial tension vector from a control polygon, replacing manual parameter selection. These tensions are then propagated according to the analytical evolution rule of the Ω-NSS scheme, preserving its theoretical properties. Numerical results show that this learned initialization strategy yields stable and consistent tension distributions, while improving reproducibility and reducing user intervention.