Mathematical Foundations of Topological Data Analysis and Its Applications to Machine Learning and Large Language Model Analysis
Informatics · 15.11.2026 · 13 (11) · 195
DOI: 10.3390/informatics13110195 ↗Kurz gefasst
Fachartikel in der Zeitschrift Informatics, online erschienen am 15.11.2026. Verfasst von Evelina Okuneva und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This review develops a mathematically structured account of topological data analysis (TDA), focusing on persistent homology and its applications to machine learning and large language model (LLM) analysis. It systematically introduces the fundamental concepts of metric and topological spaces, simplicial complexes, chain complexes and homology, filtrations, persistence modules, persistence diagrams, and stability results. Particular attention is given to the mathematical properties of Vietoris–Rips and Čech complexes, the interpretation of homology groups, and methods for comparing persistence diagrams. Applications to neural networks and LLMs are examined through geometric filtrations of embedding and activation spaces, attention-derived complexes, and zigzag persistence across transformer layers. Recent studies are critically compared in terms of their topological constructions, analyzed models and datasets, evaluation protocols, empirical findings, and computational limitations. Particular attention is given to the analysis of representation dynamics, model alignment, and hallucination detection, while distinguishing experimentally validated approaches from proposed mathematical extensions. The review also examines the influence of metric selection and filtration parameters, the computational challenges associated with large point clouds and higher-dimensional homology, and the limitations of interpreting topological characteristics in linguistic and semantic terms. Potential mathematical directions, including multiparameter persistence and functorial analysis of representations, are also discussed. The main contribution of this review is a critical mathematical synthesis connecting the theoretical foundations of persistent homology with recent developments in LLM analysis. Persistent topological features are considered complementary structural descriptors whose practical significance depends on the underlying mathematical assumptions and empirical validation, rather than universal guarantees of semantic correctness or factual reliability.
Zitieren
APA
Okuneva, E., Kochkarov, R., & Kochkarov, A. (2026). Mathematical Foundations of Topological Data Analysis and Its Applications to Machine Learning and Large Language Model Analysis. Informatics, 13(11), 195. https://doi.org/10.3390/informatics13110195
Vancouver
Okuneva E, Kochkarov R, Kochkarov A. Mathematical Foundations of Topological Data Analysis and Its Applications to Machine Learning and Large Language Model Analysis. Informatics. 2026;13(11):195. doi:10.3390/informatics13110195
BibTeX
@article{Okuneva2026,
author = {Okuneva, Evelina and Kochkarov, Rasul and Kochkarov, Azret},
title = {Mathematical Foundations of Topological Data Analysis and Its Applications to Machine Learning and Large Language Model Analysis},
journal = {Informatics},
year = {2026},
volume = {13},
number = {11},
pages = {195},
doi = {10.3390/informatics13110195}
}
EUVIMED