Researcher in Optimization
and Machine Learning

I am a researcher in applied mathematics and machine learning. Since February 2022, I am a postdoc at the University of Tübingen, working on the TRINOM-DS project led by Peter Ochs and Jalal Fadili.

Prior to that, I obtained my PhD degree at the IRIT laboratory of Toulouse under the supervision of Cédric Févotte, Édouard Pauwels and Jérôme Bolte. I was a member of the FACTORY project.

My research focuses on optimization algorithms for deep learning, and machine learning in general. I used to give tutorials and practicals at INP-ENSEEIHT where I taught probabilities, statistics and integration theory to master students.

Research interests

• Optimization: non-convex, non-smooth, stochastic, etc.
• Deep Learning
• Theory of Machine Learning and Statistics

Recent work

Inertial Newton algorithms avoiding strict saddle points

C. Castera (2021)
Preprint

arXiv

Second-order step-size tuning of SGD for non-convex optimization

C. Castera, J. Bolte, C. Févotte, E. Pauwels (2022)
Neural Processing Letters

NePL | arXiv

An inertial Newton algorithm for deep learning

C. Castera, J. Bolte, C. Févotte, E. Pauwels (2021)
Journal of Machine Learning Research

JMLR | Hal | Code

Python packages for the INNA algorithm

Pytorch Version | Keras Version | Tensorflow Version

Poster: the INNA algorithm

Poster presented at NeurIPS 2019

List of publications

• Full list

Contact