Geometric and computational complexity in bilevel optimization
17 Apr 2025, 12:00 — Room 322, UniGe DIBRIS/DIMA, Via Dodecaneso 35
![Samuel.Vaiter - [Occhiali da vista peli sul viso Occhiali]](/media/images/3mXEfCKU_400x400_2_dfaua0_qnmLiWE.width-800.jpg)
Speaker:
Samuel Vaiter
Samuel Vaiter
Abstract:
In this talk, I will introduce bilevel optimization (BO) as a powerful framework to address several machine learning-related problems, including hyperparameter tuning, meta-learning, and data cleaning. Based on this formulation, I will describe some successes of BO, particularly in a strongly convex setting, where strong guarantees can be provided along with efficient stochastic algorithms. I will also discuss the outstanding issues of this framework, presenting geometrical and computational complexity results that show the potential difficulties in going beyond convexity, at least from a theoretical perspective.
In this talk, I will introduce bilevel optimization (BO) as a powerful framework to address several machine learning-related problems, including hyperparameter tuning, meta-learning, and data cleaning. Based on this formulation, I will describe some successes of BO, particularly in a strongly convex setting, where strong guarantees can be provided along with efficient stochastic algorithms. I will also discuss the outstanding issues of this framework, presenting geometrical and computational complexity results that show the potential difficulties in going beyond convexity, at least from a theoretical perspective.
Bio:
Samuel Vaiter is currently a CNRS Research Scientist at Laboratoire J. A. Dieudonné, Université Côte d’Azur, Nice, France. He received the Ph.D. degree in applied mathematics from Université Paris-Dauphine, Paris, France, in 2014, under the supervision of Gabriel Peyré. He was a postdoctoral researcher at École Polytechnique in 2014–2015, working with Antonin Chambolle. From 2015 to 2021, he was a Research Scientist with the Institut de Mathématiques de Bourgogne, CNRS, Dijon, France. His research interests include mathematical optimization, bilevel programming, convex and nonsmooth analysis, algorithmic differentiation, and theoretical aspects of machine learning, with applications to inverse problems, signal processing, and graph-based learning.
Samuel Vaiter is currently a CNRS Research Scientist at Laboratoire J. A. Dieudonné, Université Côte d’Azur, Nice, France. He received the Ph.D. degree in applied mathematics from Université Paris-Dauphine, Paris, France, in 2014, under the supervision of Gabriel Peyré. He was a postdoctoral researcher at École Polytechnique in 2014–2015, working with Antonin Chambolle. From 2015 to 2021, he was a Research Scientist with the Institut de Mathématiques de Bourgogne, CNRS, Dijon, France. His research interests include mathematical optimization, bilevel programming, convex and nonsmooth analysis, algorithmic differentiation, and theoretical aspects of machine learning, with applications to inverse problems, signal processing, and graph-based learning.