Seminarios "La informática en la encrucijada"

¿Puede la Topología Mejorar la IA? Una Aventura con Hipergrafos y Redes Neuronales.

Pedro Real

This talk explores a new paradigm in hypergraph neural networks that fuses advanced algebraic topology with deep learning techniques. We present a dynamic framework—the scale-space topological model (s2-model)—that systematically decomposes hypergraph structures into a sequence of algebraic hypergraphs while preserving topological invariants such as the Euler-Poincaré characteristic.

Inspired by classical methods (including boundary-scale models and adaptations of the Weisfeiler-Lehman algorithm), this approach extracts novel homological and combinatorial features, dramatically enhancing the expressivity of hypergraph neural networks.

Attendees will discover how these multi-scale models can distinguish subtle structural differences, address challenges in hypergraph isomorphism, and transform analysis tasks in physical, biological, and social networks. Join us on a journey where algebra meets algorithms and topological scale unlocks new dimensions of learning power.