pipette
ESEspañol

Topological Learning under Sparsity through Huberized Distance Functionals

Tuhin Subhra Mahato, Subhra Sankar Dhar

Preprint

In the authors' words

Learning topological structure from sparsely contaminated data requires robust statistical methods, as even a small fraction of atypical observations can substantially distort the inferred topology. This article introduces a robust framework for topological learning through the Huberized Distance-to-Measure, which provides a robust filtration-generating functional for persistent homology. We establish large-sample properties of the proposed functional, and show that the associated filtration induces stable and consistent persistence diagrams. We further derive a maximum bias bound for persistence diagrams in bottleneck distance, thereby providing a quantitative measure of the sparseness associated with topological inference. The practical utility of the methodology thorough testing of statistical hypothesis problem is illustrated for a seven-circle packing configuration under sparse contamination, where it recovers the underlying topological structure. Its applicability to real data is further demonstrated through an analysis of the M-ferritin protein structure, where the method identifies a persistent feature corresponding to the iron-storage cavity.

Main resultThe abstract does not state a limitation.

Appeared: Tuesday, October 20. arXiv. Preprint, not yet peer-reviewed.