Fast Amortized Fitting of Scientific Signals Across Time and Ensembles via Transferable Neural Fields

TL;DR

This paper introduces a transferable neural field framework for efficient, scalable modeling of high-dimensional scientific signals across time and ensembles, significantly improving convergence and accuracy.

Contribution

It extends neural implicit representations to handle spatiotemporal multivariate signals and enables transfer learning across signals for faster, more accurate scientific data modeling.

Findings

Transferable features reduce reconstruction iterations by up to tenfold.

Early-stage reconstruction quality improves by multiple decibels, exceeding 10 dB in some cases.

Gradient-based physical quantities are more accurate with transferable neural fields.

Abstract

Neural fields, also known as implicit neural representations (INRs), offer a powerful framework for modeling continuous geometry, but their effectiveness in high-dimensional scientific settings is limited by slow convergence and scaling challenges. In this study, we extend INR models to handle spatiotemporal and multivariate signals and show how INR features can be transferred across scientific signals to enable efficient and scalable representation across time and ensemble runs in an amortized fashion. Across controlled transformation regimes (e.g., geometric transformations and localized perturbations of synthetic fields) and high-fidelity scientific domains-including turbulent flows, fluid-material impact dynamics, and astrophysical systems-we show that transferable features improve not only signal fidelity but also the accuracy of derived geometric and physical quantities, including density gradients and vorticity. In particular, transferable features reduce iterations to reach target reconstruction quality by up to an order of magnitude, increase early-stage reconstruction quality by multiple dB (with gains exceeding 10 dB in some cases), and consistently improve gradient-based physical accuracy.