Minimax Additive Regression under Unknown Dependent Designs
Matching minimax bounds for additive regression with unknown, dependent covariates, identifying when marginal-density estimation changes the optimal rate.
Mathematics for Interpretability
Baptiste Ferrere
Statistics, Machine Learning, Interpretability
I am a Ph.D. student in Applied Mathematics at Université de Toulouse, in collaboration with EDF R&D through a CIFRE partnership.
My research lies at the intersection of nonparametric interaction models, statistical learning theory and interpretability.
I am deeply interested in the theoretical and mathematical foundations of machine learning. I particularly enjoy developing methods grounded in statistical learning theory, functional decompositions and functional analysis, and using them to address concrete problems in explainable AI.
I study how to represent and estimate individual effects and interactions between variables, particularly when the inputs are dependent and their distribution is unknown.
My recent contributions include explicit representations for generalized functional ANOVA and minimax guarantees for additive regression under unknown dependent designs.
- Academic supervisors
- Fabrice Gamboa
Jean-Michel Loubes - Industrial supervisors · EDF
- Nicolas Bousquet
Joseph Muré
News
I am visiting Université du Québec à Montréal (UQAM) throughout October 2026, working with Arthur Charpentier and Marouane Il Idrissi on interpretability and fairness.
Our new preprint Minimax Additive Regression under Unknown Dependent Designs is available on arXiv!
Our paper on functional ANOVA for categorical inputs was accepted at ICML 2026 for an oral presentation.
Earlier news
Our new preprint Generalized Functional ANOVA: A Complete Theoretical Framework is available on arXiv!
Our new preprint Exact Functional ANOVA Decomposition for Categorical Inputs Models is available on arXiv!
Our new preprint Fourier Analysis on the Boolean Hypercube via Hoeffding Functional Decomposition is available on arXiv!
Selected publications
All publicationsGeneralized Functional ANOVA: A Complete Theoretical Framework
An explicit, distribution-adapted basis for generalized functional ANOVA with dependent continuous inputs.
Exact Functional ANOVA Decomposition for Categorical Inputs Models
A closed-form functional ANOVA for categorical models, including dependent features and non-rectangular supports.
Research directions
Statistical learning theory
Statistical guarantees for nonparametric models, including minimax rates for additive regression under unknown dependent designs.
Nonparametric interaction models
Representation and estimation of main effects and higher-order interactions using functional decompositions and generalized ANOVA.
Interpretability & UQ
Functional decompositions for model interpretability and uncertainty quantification; ongoing work on interactions in transformer architectures.
Background
Full CVBefore my Ph.D., I worked on functional ANOVA for correlated inputs during a research internship at EDF R&D. I hold an M.Sc. in Statistical Learning from Institut Polytechnique de Paris and an engineering degree from ENSAE Paris.
For research discussions and collaborations, get in touch .