Guardat en:
| Autor principal: | |
|---|---|
| Format: | Recurso digital |
| Idioma: | |
| Publicat: |
Zenodo
2025
|
| Matèries: | |
| Accés en línia: | https://doi.org/10.5281/zenodo.17455111 |
| Etiquetes: |
Afegir etiqueta
Sense etiquetes, Sigues el primer a etiquetar aquest registre!
|
Taula de continguts:
- This paper explores the geometric structure of high-dimensional vector spaces through the lens of information theory and entropy. We investigate how fundamental concepts such as Shannon entropy, typical sets, and the concentration of measure phenomenon dictate the effective geometry experienced by data distributions. By applying principles from information geometry, we characterize the space of high-dimensional probability distributions as a statistical manifold, where distances and curvatures are defined by information-theoretic metrics like the Fisher information metric. The study reveals that in high dimensions, probability mass concentrates in a thin shell, a counter-intuitive result that has profound implications for statistical inference, machine learning, and data analysis. We demonstrate that the apparent volume of these spaces is governed by entropic quantities, leading to a geometric framework where proximity and structure are naturally expressed in terms of information divergence rather than traditional Euclidean distance. This entropic perspective provides a powerful toolkit for understanding the behavior of algorithms and the intrinsic dimensionality of data in high-dimensional settings.