Skip to content

Deep Ensembles + Parametric Poisson 1D

Notebook: DeepEnsemble_ParametricPoisson1D_Tutorial.ipynb

This tutorial uses an ensemble of MLP models to learn a parametric 1D Poisson response map. Inputs are the coordinate and source parameters, output is the solution value, and uncertainty is visualized as a band that widens on OOD forcing settings. The notebook now explicitly distinguishes this parameter-space OOD effect from a separate physical-space sparse-data story.

Primary references:

  • Lakshminarayanan et al. (2017), Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
  • Rasmussen & Williams (2006) for the broader uncertainty calibration context