Date: Wed. July 9, 2025 (11:30H)
Event: DeustoCCM Seminar

11:30H. Session 01
Title: Robustness with neural ordinary differential equations
Speaker: Prof. Rafael Orive Illera
Affiliation: UAM, Autonomous University of Madrid

Abstract. The robustness of a machine learning refers to how the model responds to data disturbances, which can be random, associated with measurement errors, or due to targeted and malicious disturbances. Neural differential equations have emerged as a suitable tool both for developing learning models and for interpreting the behavior of neural networks within them. Using optimal differential equation control techniques, we study the problem of robustness and new learning models.

 
12:15H. Session 02
Title: New deep learning models and perspectives for continuous-time glucose monitoring
Speaker: Antonio Álvarez López
Affiliation: UAM, Autonomous University of Madrid

Abstract. In this talk, I will present recent progress from my doctoral thesis on modeling the dynamics of probability distributions using time series data. This problem appears in various areas, notably digital health, where understanding how biomarkers like glucose evolve over time is key to studying chronic diseases such as diabetes. Our approach uses a Gaussian mixture model that represents the evolution of a probability density. We combine Maximum Mean Discrepancy (MMD) for Gaussian parameter estimation at each time, with a neural ODE governing the evolution of mixture weights over time. The resulting model is interpretable, computationally efficient, and sensitive to subtle distributional shifts. Simulations show our method achieves accuracy comparable to current state-of-the-art methods. We illustrate its effectiveness with data from a digital clinical trial, demonstrating how interventions influence glucose distributions over time.
 

WHERE
University of Deusto, DeustoTech.
Room: Logistar
Unibertsitate Etorb., 24, 48007 Bilbao, Bizkaia