Efficient and robust persistent homology for measures (Q340536)

One of the tools used in topological data analysis is the persistent homology and the most common goal in the analysis is to apply the persistence algorithm to distance functions. In the studies on these problems there are two distinct research directions: (1) the distance to the data set is replaced with a distance to a measure induced by that data set, (2) the use of sparse filtrations. In the paper, the authors combine these two directions together by combining the robustness of the distance to a measure with the efficiency of sparse filtrations.NEWLINENEWLINEAt the beginning the authors prove that if we have two close probability measures, then the persistence diagrams of the sublevel sets filtration of their distance to measure functions are close. The result is proved in triangulable metric spaces, i.e., metric spaces that are homeomorphic to a locally finite simplicial complex. Because computing the persistence diagram of the sublevel sets filtration of the distance to a measure requires knowing the sublevel sets, that are not generally easy to compute, the authors propose an approximation paradigm for the distance to a measure that replaces the sublevel set by a union of balls. The method uses \(O(n)\) balls for inputs of \(n\) samples and works in any metric space. Next, the authors introduce a generalization of the rips filtration -- the weighted rips filtration. The proposed construction allows to approximate in some cases the persistence diagram of the distance to a measure. Because the weighted rips filtration is too large to construct in full for large instances the authors propose construction of a sparse approximation that has linear size in the number of points. Finally, some examples are presented. The examples present the quality of the approximation, the stability of the diagrams with respect to noise and the size of the filtration after sparsification.