Robust Barycenters of Persistence Diagrams

Keanu Sisouk, Eloi Tanguy, Julie Delon, Julien Tierny · 2025

Read Robust Barycenters of Persistence Diagrams by Keanu Sisouk, Eloi Tanguy, Julie Delon, Julien Tierny (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-14904.txt.

Introduction

Keanu Sisouk, Eloi Tanguy, Julie Delon, and Julien Tierny compute averages of persistence diagrams that resist outliers in a 2025 topological data analysis paper. Persistence diagrams summarize the shape of data as collections of points, and their averages, called barycenters, are usually built from optimal transport plans. The classical method only handles the squared Wasserstein distance.

The authors adapt a fixed-point method to general transport costs, especially costs between the first and second power that shrug off outliers. Two uses show the benefit: clustering diagrams within their own metric space, and building dictionary encodings of diagram collections. In both, the generalized barycenter proves more robust than the standard one. A public Python implementation accompanies the work, so others can apply the robust averaging directly.

Catalog record

CREATORKeanu Sisouk, Eloi Tanguy, Julie Delon, Julien Tierny
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-14904

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Robust Barycenters of Persistence Diagrams Keanu Sisouk, Eloi Tanguy, Julie Delon, Julien Tierny Abstract This short paper presents a general approach for computing robust Wasserstein barycenters [ 2 , 78 , 79 ] of persistence diagrams. The classical method consists in computing assignment arithmetic means after finding the optimal transport plans between the barycenter and the persistence diagrams. However, this procedure only works for the transportation cost related to the q q - Wasserstein distance W q W_{q} when q = 2 q=2 . We adapt an alternative fixed-point method [ 74 ] to compute a barycenter diagram for generic transportation costs ( q > 1 q>1 ) , in particular those robust to outliers , q ∈ ( 1 , 2 ) q\in(1,2) . We show the utility of our work in two applications : (i) the clustering of persistence diagrams on their metric space and (ii) the dictionary encoding of persistence diagrams [ 71 ] . In both scenarios, we demonstrate the added robustness to outliers provided by our generalized framework. Our Python implementation is available at this address: https://github.com/Keanu-Sisouk/RobustBarycenter . I Introduction With measurement devices and numerical techniques being more and more precise, the resulting datasets have become more and more complex geometrically. This complexity hinders the users during any exploration and analysis to interpret them. Those challenges motivate the conception of expressive, informative , concise and simple data abstractions, encoding the main features and patterns of interest of the data. Topological Data Analysis (TDA) [ 25 ] is a family of tools developed to address this challenge. It aims to provide simple objects, called topological representations, describing the main topological structures in a dataset . TDA has become a staple for analyzing scalar data, its efficiency and robustness being shown in numerous visualization tasks [ 41 ] . It has proven to be successful in many applications, several examples include turbulent combustion [ 11 , 37 , 49 ] , material sciences [ 29 , 39 , 40 , 72 ] , nuclear energy [ 52 ] , fluid dynamics [ 44 , 56 ] , bioimaging [ 3 , 9 , 16 ] , chemistry [ 6 , 34 , 57 , 58 ] or astrophysics [ 70 , 73 ] to name a few. Among the different topological representations, such as the merge and contour trees [ 75 , 14 , 51 , 1 , 15 , 35 ] , the Reeb graph [ 7 , 61 , 76 , 60 , 24 , 36 ] , or the Morse-Smale complex [ 33 , 27 , 26 , 10 , 22 , 66 , 69 , 38 ] , the Persistence Diagram ( Fig. 1 ) as been prominently used and studied. It is a simple and concise topological representation, which encodes the main topological features of data. With the increase of geometrical complexity discussed above, a new challenge has come forth as users are confronted to the emergence of ensemble datasets . With these representations, a given phenomenon is not only described by a single dataset, but by a collection of datasets, called ensemble members . This results in the analysis of an ensemble where each element is a topological representation, such as a persistence diagram . However, this process moves the problem of analyzing an ensemble of scalar fields to analyzing an ensemble of persistence diagrams. To address this challenge , the question of finding a good representative of such an ensemble has emerged. An established answer to that question is the notion of Wasserstein barycenter [ 2 , 78 ] of persistence diagrams, based on the so called q - Wasserstein distance [ 81 ] . To compute such barycenter s , several algorithms have been proposed for q=2 [ 78 , 48 , 80 ] . This barycenter emulates the behavior of an average , and as such , it is also sensitive to outliers. Works on Wasserstein medians counter this issue [ 77 , 13 ] , showing their stability properties. In the context of probability measures, extensions to generic transportation costs have been introduced recently [ 12 , 74 ] . This short paper proposes to leverage generic transportation costs for computing robust q -Wasserstein barycenters of ensembles of persistence diagrams , with q>1 . For this, we adapted a recent fixed-point method [ 74 ] from generic probability measures to persistence diagrams. As discussed above, the resulting barycenters have the advantage of being more robust to the presence of outlier diagrams in the ensemble. Experiments on synthetic and real-life data showcase this property for typical ensembles of persistence diagrams used previously in the literature. We show the utility of our barycenters by using them in a clustering problem on the Wasserstein metric space and by using them for a Wasserstein dictionary encoding problem [ 71 ] . We give details on this robust barycentric framework in Sec. III , then we show the utility of these robust barycenters with experiments in Sec. IV with two applications in Sec. IV-A and Sec. IV-B . I-A Related work The literature related to our work can be classified into two main categories, reviewed in the following: (i) ensemble visualization, and (ii) topological methods for ensemble of persistence diagrams. (i) Ensemble visualization: A common approach to model data variability is using ensemble datasets. In this setting, the variability is modeled by a succession of empirical observations (i.e ., the members of the ensemble). Existing approaches compute geometrical objects encoding the features of interest (such as level sets and streamlines for example) for each member of the ensemble. Then, a representative of this ensemble of geometrical objects can be estimated. In light of this, several techniques has been established. For example, spaghetti plots [ 23 ] are used to study level-set variability, especially for weather data [ 65 , 68 ] . Box-plots [ 82 , 53 ] are used to analyze the variability of contours and curves. In the case of flow ensemble, Hummel et al. [ 42 ] has proposed a Lagrangian framework for classification purposes. More precisely, clustering techniques have been studied to highlight the main trends in ensemble of streamlines [ 30 ] and isocontours [ 31 ] . However, only few approaches applied those strategies to topological representations. Favelier et al. [ 28 ] and Athawale et al. [ 4 ] introduced techniques for analyzing the geometrical variability of critical points and gradient separatrices respectively. Ensemble layouts have been proposed for contour trees [ 47 , 84 ] . However, the above techniques do not focus on the computation of a representative of an ensemble of topological objects. (ii) Topological methods: To find a representative of an ensemble of persistence diagrams, notions from optimal transport [ 54 , 43 ] were adapted to persistence diagrams. A central notion is the so-called Wasserstein distance [ 67 ] . The Wasserstein distance between persistence diagrams [ 25 ] ( Sec. II-B ) has been studied by the TDA community [ 18 , 19 ] . This distance is computed by solving an assignment problem, for which exact [ 55 ] and approximate [ 5 , 45 ] implementation can

Read the complete text