Théophile Cantelobre

Research Assistant @ SIERRA Team (Inria Paris)

I am interested in theoretical guarantees for machine learning methods: proving they are reliable and efficient… sometimes that they are optimal!

Currently, I am working between Inria Paris and London under the direction of Alessandro Rudi, Benjamin Guedj, and Carlo Ciliberto.

I studied Mathematics & Engineering in a dual masters program between Mines ParisTech (Cycle Ingénieur Civil) and Sorbonne Université (M2A) in Paris, France.

In the past, I’ve worked on PAC-Bayes guarantess for structured prediction at Inria & UCL, and state estimation for underwater robotics at Schlumberger-Doll Research.

Learn more…


Dec 7, 2020 New preprint with Benjamin Guedj, María Pérez-Ortiz, and John-Shawe Taylor is online.
Oct 25, 2020 IROS 2020 is open for free to everyone, come check out our work!


  1. A PAC-Bayesian Perspective on Structured Prediction with Implicit Loss Embeddings
    arXiv 2012.03780, 2020
  2. A real-time unscented Kalman filter on manifolds for challenging AUV navigation
    Théophile Cantelobre, Clément Chahbazian, Arnaud Croux, and Silvère Bonnabel
    In IEEE IROS, 2020