Welcome to Thibaut Germain’ webpage!
About me
My research interests lie at the intersection of machine learning, geometry, and dynamical systems. More specifically, I focus on developing machine learning methods tailored to dynamical systems and time series, with a particular emphasis on the interpretability, comparison, and transport of dynamic behaviors.
Since March 2025, I have been a postdoctoral researcher at Centre de Mathématiques Appliquées de Polytechnique (CMAP), working with Karim Lounici and Rémi Flamary on domain adaptation for stochastic dynamical systems through their transfer operators.
Before, I was PhD student at Centre Borelli, a research lab from ENS Paris-Saclay under the supervision of Charles Truong and Laurent Oudre. I developed shape-based methods tailored for the discovery and statistical analysis of time series patterns with a particular focus on biomedical applications [thesis pdf].
contact: thibaut.germain.pro [at] gmail.com
News
- February 2026: Paper accepted on ICLR on novel Wasserstein metrics between operator representation of dynamical systems.
- May 2025: Paper accepted to ICML on the hard-coding of time series invariances in convolutional neural layers to improve robustness and generalization.
- April 2025: Survey paper accepted to Very Large DataBases journal (VLDB) on the problem of motif discovery in time series.
reference
- Guerrini, V., Germain, T., Truong, C., Oudre, L, & Boniol, P. (2025). Time Series Motif Discovery: A Comprehensive Evaluation. Proceedings of the VLDB Endowment. [python package][github][pdf]
- September 2024: Paper accepted to NeurIPS on the representation of time series through diffeomorphic deformations.
- July 2024: I have been to Duke University as a visiting researcher in Sapiro lab to work on facial emotions recognition and analysis for therapeutic evaluation of eating disorders.