Sheaf-theoretic stratification learning from geometric and topological perspectives (Q2022620)

Brown and Wang present a new algorithm for learning decompositions of topological spaces, motivated by the concept of stratifications from intersection homology, in which a topological space is decomposed into different strata. Taking a more general perspective, the authors develop a new approach for inferring a decomposition of input data into disjoint subsets, each of which capturing structural information about the whole data set. This is achieved by constructing a sheaf structure over such spaces. Brown and Wang then go on to show that such sheaf-driven stratifications exist and, under certain conditions, are even unique to a space. In addition, the paper presents an algorithm for obtaining such stratifications for Kolmogorov spaces (\(T_0\) spaces), thus also enabling calculations in the setting of simplicial complexes. Finally, the paper also contains several worked examples of the proposed stratification method in practice, linking them to the construction of `Mapper' graphs.