Optimizing persistent homology based functions - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2021

Optimizing persistent homology based functions

Mathieu Carriere
Frédéric Chazal
Marc Glisse


Solving optimization tasks based on functions and losses with a topological flavor is a very active,growing field of research in data science and Topological Data Analysis, with applications in non-convexoptimization, statistics and machine learning. However, the approaches proposed in the literatureare usually anchored to a specific application and/or topological construction, and do not come withtheoretical guarantees. To address this issue, we study the differentiability of a general map associatedwith the most common topological construction, that is, the persistence map. Building on real analyticgeometry arguments, we propose a general framework that allows us to define and compute gradientsfor persistence-based functions in a very simple way. We also provide a simple, explicit and sufficientcondition for convergence of stochastic subgradient methods for such functions. This result encompassesall the constructions and applications of topological optimization in the literature. Finally, we provideassociated code, that is easy to handle and to mix with other non-topological methods and constraints, aswell as some experiments showcasing the versatility of our approach.
Fichier principal
Vignette du fichier
_ICML21__Optimizing_persistent_homology_based_functions.pdf (1.03 Mo) Télécharger le fichier
Vignette du fichier
pc1000.png (37.79 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Format : Figure, Image

Dates and versions

hal-02969305 , version 1 (19-10-2020)
hal-02969305 , version 2 (18-02-2021)



Mathieu Carriere, Frédéric Chazal, Marc Glisse, Yuichi Ike, Hariprasad Kannan. Optimizing persistent homology based functions. ICML 2021 - 38th International Conference on Machine Learning, Jul 2021, Virtual conference, United States. pp.1294-1303. ⟨hal-02969305v2⟩
287 View
767 Download



Gmail Facebook X LinkedIn More