Sets of exemplary microstructure-property data generated via active learning and numerical simulations

Abstract

This publication contains three exemplary data sets generated via active learning and numerical simulations. The active learning approach used is query-by-committee. For comparison, data is also generated using classical sampling approachs. The first data set originates from a toy example that is based on an appoximated Dirac delta function, for which data was generated randomly and via query-by-committee. The second example is part of a parameter identification problem in materials modeling, for which data was generated via Latin Hypercube design, a knowledge-based approach and query-by-committee. The third example is about generating artificial bcc rolling textures, for which data was generated via Latin Hypercube design, query-by-committee and an extended query-by-committee approach that prevents sampling in regions out of scope.