Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 245 - SLDS CSpeed 4
Type: Contributed
Date/Time: Wednesday, August 11, 2021 : 10:00 AM to 11:50 AM
Sponsor: Section on Statistical Learning and Data Science
Abstract #319090
Title: Robust Persistence Diagrams Using Reproducing Kernels
Author(s): Siddharth Vishwanath* and Kenji Fukumizu and Satoshi Kuriki and Bharath Sriperumbudur
Companies: Penn State University and Institute of Statistical Mathematics and Institute of Statistical Mathematics and Penn State University
Keywords: Topological Data Analysis; Persistent Homology; Robust Statistics; Robust Topological Inference; Reproducing Kernels; RKHS

Persistent homology has become an important tool for extracting geometric and topological features from data, whose multi-scale features are summarized in a persistence diagram. From a statistical perspective, however, persistence diagrams are very sensitive to perturbations in the input space. The central objective of our work is to develop outlier robust persistence diagrams for topological data analysis. We do so by employing robust density estimators constructed using reproducing kernels. Using an analogue of the influence function on the space of persistence diagrams, we investigate the sensitivity of persistence diagrams from filtrations induced by several functions (e.g., KDE, robust KDE, DTM) and demonstrate the robustness of the proposed method. The robust persistence diagrams are also shown to be consistent estimators in bottleneck distance, with the convergence rate controlled by the smoothness of the kernel, i.e., under some appropriate choices, it is possible to construct outlier-robust persistence diagrams without compromising on statistical efficiency. Finally, we demonstrate the performance of the proposed approach in machine learning applications.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program