Sequential Conditional Independence Testing with Machine Learning Models
Reyero-Lobo, A., Meziu, M., Arias, S. U., and Grünwald, P. (2026). Sequential conditional independence testing with machine learning models.
CV
Education
- PhD in Machine Learning (started in October 2024)
- Institut de Mathématiques de Toulouse and MIND, Inria Paris-Saclay
- Under the supervision of Pierre Neuvial and Bertrand Thirion
- ELLIS PhD student with Peter Grünwald as second supervisor
- MSc in Mathematics and Artificial Intelligence (September 2022-October 2024)
- Institut de Mathématiques d’Orsay & CentraleSupélec, Université Paris-Saclay, 18.43/20
- Double degree in Mathematics and Computer Science (Septembre 2019-Septembre 2022)
- First and second year in Universidad Complutense de Madrid, 9.09/10
- Third year in Université Paris-Saclay, 17.9/20
Work experience
- Research exchange in Hypothesis Testing (01/01/2026-30/06/2026)
- Centrum Wiskunde & Informatica (CWI), Amsterdam
- Under the supervision of Peter Grünwald
- Sequential Conditional Independence Testing with Machine Learning models using e-values.
- Preprint pdf
- Research in Machine Learning (01/10/2024-01/10/2027)
- Institut de Mathématiques de Toulouse and MIND, Inria Paris-Saclay
- Under the supervision of Pierre Neuvial and Bertrand Thirion
- Scientific Discovery through Conditional Independence Testing with Model-X framework and Interpretable Machine Learning.
- Research in Machine Learning (01/04/2024-30/09/2024)
- Institut de Mathématiques de Toulouse and MIND, Inria Paris-Saclay
- Under the supervision of Pierre Neuvial and Bertrand Thirion
- Theoretical study of multiple model-agnostic variable importance measures and the use of the knockoff framework for controlled variable selection.
- Research in Statistical Learning (01/04/2023-31/07/2023)
- Laboratoire de Probabilités, Statistique, et Modélisation (LPSM), Sorbonne Université.
- Under the supervision of Alexis Ayme, Claire Boyer, Erwan Scornet and Aymeric Dieuleveut.
- Study of the optimality of linear classifiers in the context of prediction with missing data: adaptability to the pattern-by-pattern decomposition.
- AISTATS 2025 pdf.
- Internship report pdf.
- Research in Optimization/ Operational Research (01/06/2022-31/07/2022)
- Laboratoire Interdisciplinaire des Sciences Numériques, Université Paris-Saclay.
- Under the supervision of Nicolas Dupin
- Use of linear programming and insertion algorithms to define neighborhoods and obtain local optimums. Ultimately, perturb and modify neighborhoods to enhance solutions.
- Published in Roadef 2024 pdf.
Scholarships
- Mobility Grants to visit Peter Grünwald 2026
- ELSA Mobility Grant 2025
- ELSA.
- G-Research Mobility Grant 2024
- Inspiring Minds Experts Mentoring 2024 (With Mireille Chaleyat-Maurel)
- Excellence scholarship MixtAI 2023
- Institut DataIA, Université Paris-Saclay.
- Scholarship for young researchers for the 10th HLF (Sept. 2023)
- Excellence scholarship 2019, 2020, 2021 and 2022
- Science, University and Innovation Council of Madrid.
- Highest Honors with Jury Congratulations(High school) 2019
- Colegio Valdeluz, Madrid
Publications
Aggregate Models, Not Explanations: Improving Feature Importance Estimation
Joseph Paillard, Angel Reyero Lobo, Denis A. Engemann, Bertrand Thirion. Aggregate Models, Not Explanations: Improving Feature Importance Estimation. In Proceedings of the International Conference on Machine Learning (ICML), 2026.
Semi-knockoffs: a model-agnostic conditional independence testing method with finite-sample guarantees
Angel Reyero-Lobo, Bertrand Thirion, and Pierre Neuvial. Semi-knockoffs: a model-agnostic conditional independence testing method with finite-sample guarantees. In Proceedings of the International Conference on Machine Learning (ICML), 2026.
A principled approach for comparing Variable Importance
Reyero-Lobo, A., Neuvial, P. , & Thirion, B. (2025). A principled approach for comparing Variable Importance.
When Knockoffs fail: diagnosing and fixing non-exchangeability of Knockoffs
Blain, A., Reyero Lobo, A., Linhart, J., Thirion, B., & Neuvial, P. (2025). When Knockoffs fail: Diagnosing and fixing non-exchangeability of Knockoffs.
Conditional Feature Importance revisited: Double Robustness, Efficiency and Inference
Reyero Lobo, A., Neuvial, P., and Thirion, B. Sobol-cpi: a doubly robust conditional permutation importance statistic. 2025
Measuring Variable Importance in Heterogeneous Treatment Effects with Confidence
Paillard, J., Reyero Lobo, A., Kolodyazhniy, V., Thirion, B., & Engemann, D. A. (2025). Measuring variable importance in heterogeneous treatment effects with confidence.
A primer on linear classification with missing data
Reyero Lobo, A. D., Ayme, A., Boyer, C., and Scornet, E. (2025). A primer on linear classification with missing data.
Analyse statistique d’hill climbers à voisinage large pour le problème d’ordonnancement linéaire
Angel David Reyero Lobo, Nicolas Dupin. Analyse statistique d’hill climbers à voisinage large pour le problème d’ordonnancement linéaire. ROADEF 2024, Mar 2024, Amiens, France. ffhal-04450707
Talks
Semi-knockoffs: a model-agnostic Conditional Independence Testing method with finite-sample guarantees
Talk at International Biometrics Conference 2026, Seoul, South Korea
Sequential Conditional Independence Testing with Machine Learning Models
Poster at ICML 2026 Hypothesis Testing Workshop, Seoul, South Korea
Semi-knockoffs: a model-agnostic Conditional Independence Testing method with finite-sample guarantees
Poster at ICML 2026, Seoul, South Korea
Sequential Conditional Independence Testing with Machine Learning
Poster at SAVI 2026, University of Twente, Netherlands
Semi-knockoffs: a model-agnostic Conditional Independence test robust to Model-X misspecification
Talk at Big Stats seminar, Amsterdam UMC, Netherlands
Beyond Coefficients: Understanding Variable Importance in Modern ML
Tutorial at COmputation - MEthodology - ECONometrics, Amsterdam, Netherlands
Conditional Feature Importance: Double Robustness, Efficiency and Inference
Talk at SIAM Conference on Uncertainty Quantification, Minneapolis, USA
A principled approach for comparing Variable Importance
Talk at Workshop: Methods for Explainable Machine Learning in Health Care, Amsterdam, Netherlands
A principled approach for comparing Variable Importance
Talk at VITE2025 : Explanability for high-dimensional statistics, Montpellier, France
Sobol-CPI: a Doubly Robust Conditional Permutation Importance Statistic
Talk at 56th Journées de Statistique de la SFdS, Marseille, France
A primer on classification with missing values
Talk at 56th Journées de Statistique de la SFdS, Marseille, France
When knockoffs fail: diagnosing and fixing non-exchangeability of knockoffs
Talk at Invited speaker at the 10th Channel Network Conference, Liège, Belgium
A primer on linear classification with missing data
Poster at 28th International Conference on Artificial Intelligence and Statistics (AISTATS), Phuket, Thailand
Robust-CPI: A Double Robust Approach to improve Variable Selection.
Talk at Workshop on statistical inference for complex data, Alan Turing building, Inria Saclay, France
Inferring variable importance and controlled variable selection
Talk at Second year of MSc in Mathematics and AI internship presentation, Institut de Mathémathiques d'Orsay, Université Paris-Saclay, France
Problematic of inferring variable importance with high-dimensional and correlated settings
Poster at Poster session, ELLIS Doctoral Symposium 2024, Paris, France
Prediction with missing values and linear classifiers
Talk at First year of MSc in Mathematics and AI internship presentation, Institut de Mathémathiques d'Orsay, Université Paris-Saclay, France