The Surprising Effectiveness of Noise Pretraining for Implicit Neural Representations

ArXi:2603.29034v1 Announce Type: new The approximation and convergence properties of implicit neural representations (INRs) are known to be highly sensitive to parameter initialization strategies. While several data-driven initialization methods nstrate significant improvements over standard random sampling, the reasons for their success -- specifically, whether they encode classical statistical signal priors or complex features -- remain poorly understood. In this study, we explore this phenomenon through a series of experimental analyses leveraging noise pre.