Egilea: Jose Luis Flores Barroso

Izenburua: Advances in supervised discretization in machine learning

Zuzendariak: Borja Calvo / Aritz Pérez

Eguna: 2026ko uztailaren 17an
Ordua: 10:30h
Lekua: Ada Lovelace aretoa (Informatikako fakultatea)

Abstract:

"The principal aim of this dissertation is to overcome the limitations of current discretization algorithms in modern machine learning by introducing three novel density-based approaches: first, a kernel density method that efficiently estimates conditional class probabilities to provide native robustness against labeling errors; second, an extension into weakly supervised learning capable of handling probabilistic labels for the first time; and finally, a privacy-preserving distributed method designed for federated learning, which computes global discretization policies at a server level using only basic local statistics, thereby protecting user data and optimizing computational resources."