Gaussian Process Regression Applied to Atom Probe Tomography Data Reconstruction

We propose Gaussian process regression, a flexible Bayesian inference approach, as a method to enhance atom probe tomography data reconstruction. Simulated evaporation of single-crystal and polycrystalline specimens, incorporating surface diffusion effects, was reconstructed using radial basis function and Matérn kernels. Comparison with the standard wide field-of-view reconstruction algorithm shows that both kernels outperform traditional geometry-based reconstruction, with the Matérn kernel proving more effective for long-range periodic datasets. This work highlights the potential of Bayesian inference in atom probe tomography reconstruction and identifies realistic pathways toward its application to experimental data.