Date: December 1 – 2, 2025
Event: Control, Inverse Problems and Machine Learning – Third meeting of the COPI2A network
Talk: Spiking Neural Networks: theoretical framework for universal approximation and training
Speaker: Umberto Biccari. DeustoCCM, University of Deusto
Slides
Summary: this work provides a rigorous mathematical analysis of Spiking Neural Networks based on Leaky Integrate-and-Fire neurons. It proves a universal approximation theorem, establishing their expressive power, and studies the well-posedness and stability of spike dynamics across layers. The results clarify how spike counts evolve and identify conditions leading to stable, decreasing, or resonant spiking behavior, offering theoretical guarantees for the use of SNNs in Machine Learning.
DeustoCCM participated in the meeting as a member of the COPI2A network, representing DeustoTech. The meeting offered an opportunity of maintaining contacts with the other participating agents, monitor recent advances in Control, Inverse Problems, and Machine Learning, and strengthen collaborations within the community.
_
COPI2A (Control, Problemas Inversos y Aprendizaje Automático) is a Spanish network on Control and Inverse Problems of systems governed by ODEs and PDEs and Machine Learning. Deusto Tech is represented by the Chair of Computational Mathematics.