Cédric Rommel
Chief Machine Learning Scientist at Ava
I am a machine learning scientist and tech lead of the AI team at Ava, working in speaker recognition research to help breaking communication barriers between the deaf & the hearing world. My current research and engineering topics include mainly real-time speaker diarization with multiple channels, metric learning and neural voice embeddings.
I obtained my PhD (CIFRE) in Machine Learning and Optimal Control from the Ecole Polytechnique, under the supervision of Frédéric Bonnans and Pierre Martinon from INRIA's COMMANDS team. I was also greatly advised by the researchers from the French start up Safety Line, which funded my PhD. During my thesis I could develop practical solutions for identifying civil aircraft dynamical models using historical flights data, and use such models to reduce future flights fuel consumption and CO2 emissions with model-based reinforcement learning. My work is now part of the product OptiClimb, which is currently being used to deliver fuel efficient trajectories for flights all over the globe.
I also hold a MSc. in engineering and applied mathematics from MINES ParisTech.
You can find my CV here.
My main research interests are:
- Main fields of application: Speaker Recognition, Trajectory Optimization, System Identification.
- Reinforcement Learning: Model-based (optimal control, HJB, dynamic programming) and model-free (yet unexplored).
- Supervised Learning: RNNs, Functional data, Multi-task/Multi-output learning, feature selection, kernel methods.
- Unsupervised Learning: Density estimation, Online clustering.
- Optimization: First order methods, nonsmooth nonconvex optimization, alternating minimization.
Reviewing work
NeurIPS 2019, ICML 2020, ...
Ph.D. Thesis
- C. Rommel. Data analysis for aircraft trajectory optimization (Exploration de données pour l’optimisation de trajectoires aériennes), 2018. (manuscript, slides)
Preprints
- C. Rommel, J. F. Bonnans, B. Gregorutti, P. Martinon. Structured Feature Selection of Continuous Dynamical Systems for Aircraft Dynamics Identification, 2018. (preprint)
- C. Rommel, J. F. Bonnans, B. Gregorutti, P. Martinon. Quantifying the Closeness to a Set of Random Curves via the Mean Marginal Likelihood, 2018. (preprint)
- C. Rommel, J. F. Bonnans, B. Gregorutti, P. Martinon. Block sparse linear models for learning structured dynamical systems in aeronautics, 2017. (preprint)
Publications
- C. Rommel, J. F. Bonnans, B. Gregorutti, P. Martinon. Gaussian Mixture Penalty for Trajectory Optimization Problems IAAA Journal of Guidance, Control, and Dynamics, 2019. (article, preprint)
- C. Rommel, J. F. Bonnans, B. Gregorutti, P. Martinon Aircraft dynamics identification for optimal control. 7th European Conference on Aeronautics and Space Sciences (EUCASS), 2017. (article)
Poster
- June 25-26th 2018 • Paris • Data Science Summer School, Ecole Polytechnique • Quantifying the Closeness to a Set of Random Curves via the Mean Marginal Likelihood. (pdf here)
- September 14-15th 2017 • Paris • Junior Conference on Data Science and Engineering, Université Paris-Saclay • Aircraft Dynamics Identification. (pdf here)
- August 28th - September 1st 2017 • Paris • Data Science Summer School, Ecole Polytechnique • Aircraft Dynamics Identification. (pdf here)
- October 20-21th 2016 • Paris • Mathématiques, Oxygène du numérique, IHP • Exploration de données pour l'optimisation de trajectoires aériennes. ⇦Awarded best poster price !!
- October 4-5th 2016 • Paris • Workshop Transition Energetique, EDF • Exploration de données aéronautiques au service de la consommation de carburant.
Talks
- November 21st 2018 • Paris • PGMO Days 2018 • Aircraft trajectory optimization under unknown dynamics. (slides)
- August 28th 2018 • Iasi • COMPSTAT 2018 • Quantifying the Closeness to a Set of Random Curves via the Mean Marginal Likelihood. (slides)
- July 2nd 2018 • Bordeaux • ISMP 2018 • Gaussian Mixture Penalization for Trajectory Optimization Problems. (slides)
- November 14th 2017 • Paris • PGMO Days 2017 • Multi-task Bolasso based aircraft dynamics identification. (slides)
- October 4th 2017 • Paris • CMAP 1st day meeting 2017 • Parametric estimator of civil aircraft dynamics.
- July 4th 2017 • Milan • EUCASS 2017 • Aircraft dynamics identification for optimal control.
- May 5th 2017 • Paris • CMAP PhD Seminar • Aircraft Dynamics Identification
- February 23rd 2017 • Paris • CMAP Journal Club • Presentation of: C. Ciliberto, A. Rudi, L. Rosasco. A consistent regularization approach for structured prediction. NIPS, 2016 (slides)