Dimensionality reduction
AEVA needs an operation to compute a medial axis or some reduced-dimensionality representation of a source model.
The medial axis is known to have stability issues. Something more robust to noise might be an approach like Laplace-Beltrami Eigenvalues and Topological Features of Eigenfunctions for Statistical Shape Analysis which computes a reeb graph (single-parameter curve) after removing noise via persistence-diagram simplification. An example of its use on combustion simulation data shows how it not only defines a principal axis, but also a series of surface curves corresponding to "distance along the principal axis" that can be useful in feature characterization and modeling.
Edited by David Thompson