Évènement

Luc Brogat-Motte
Istituto Italiano di Tecnologia
https://lmotte.github.io. https://lmotte.github.io.

Date(s) : 22/09/2025   iCal
14h00 - 16h00

This talk will present recent contributions and ongoing directions in supervised learning for non-i.i.d. and structured data. It will present algorithms that are both efficient in practice and come with formal performance guarantees. These approaches are available as open-source software and have been validated on both synthetic benchmarks and real-world datasets.

The first part will focus on structured prediction methods designed to overcome the curse of the output dimension, providing finite-sample learning bounds and computational complexity guarantees, and demonstrating significant empirical improvements over state-of-the-art approaches in applications such as metabolite identification.

The second part will address learning from time-dependent data, combining statistical learning theory with stochastic calculus to develop estimation methods for continuous-time, nonlinear stochastic dynamical systems. Particular attention will be given to stochastic differential equation estimation and to safe learning for controlled systems under uncertainty, with applications in areas such as robotics and biology.

Emplacement
Saint-Charles - FRUMAM (2ème étage)

Catégories

• Séminaire