Séminaire de Probabilités et Statistique
Monday 15 June 2026 à 13:45 - UM - Bât 09 - Salle 430 (4e étage)
Mathis Deronzier (IMT)
Gaussian Process Regression under Monotonicity and Fairness Constraints
This presentation will introduce the framework of constrained Gaussian Process (cGP) Regression, beginning with known results in the GP literature and the notion of the Maximum A Posteriori (MAP) predictor — the central object of the cGP theory — before turning to algorithmic methods for constructing these predictors. In this context, we will briefly present the generalization of cGP to the block-additive setting, i.e., sums of functions defined on subsets of variables. The second part of the talk will focus on fairness constraints, namely statistical parity and no disparate treatment. We will present the construction of a fair GP, the extraction of a MAP predictor and develop its properties.
ATTENTION : Séminaire exceptionnellement en salle 430 (4e étage du bâtiment 9), également retransmis sur zoom : https://umontpellier-fr.zoom.us/j/99660917688