Predicting the Unknown: Smarter Uncertainty Estimates for Physics-Based AI

A new approach focuses uncertainty quantification in neural operator models on key structural components, improving the accuracy and efficiency of predictions for complex physical systems.
![As a system approaches an electronic quantum critical point, phonon softening and renormalization of the long-wavelength dispersion increase the density of low-energy phonon states, with the thermally excited phonon phase-space volume scaling as [latex]T^{d/z_p}[/latex], where [latex]d[/latex] represents spatial dimension and [latex]z_p[/latex] is the phonon dynamical exponent-a larger [latex]z_p[/latex] indicating a greater low-energy phonon phase space and, consequently, increased scattering potential, particularly when [latex]z_p > d[/latex].](https://arxiv.org/html/2603.11176v1/x1.png)