Florentin Coeurdoux
👋Hi, I’m a PhD Student on a quest to mechanize the scientific method by building artificial intelligence systems.
This involves working on Bayesian inference, Deep Learning, Generative models and Optimal Transport with the goal to effectively apply these to domain sciences.
About
I started my PhD at ANITI (Toulouse INP) within the SC group of the IRIT laboratory on November 2020, under the supervision of Nicolas Dobigeon and Pierre Chainais . Before this, I graduated from an MEng at ENSAI in applied mathematics and computer science, and from the MBA master program of IGR in financial engineering.
In Autumn/Winter 2022/2023, I was a research visiting scholar at the Department of Statistics of the University of Oxford working with Arnaud Doucet.
Previously, I worked as a Software/Research Engineer at Credit Mutuel, leveraging machine learning to build better risk management systems. I am currently a member of the scientific advisory board for data science at AssessFirst
Research Interests
Deep Generative models
- Nomalizing Flows: constructing complex distributions by transforming a probability density through a series of invertible mappings.
- Diffusion based model: define a Markov chain of diffusion steps to slowly add random noise to data and then learn to reverse the diffusion process to construct desired data samples from the noise.
- Generative Adversarial Network: implicit generative modeling algorithms.
Statiscial Inference
- Density estimation: estimating the probability density function of the population from the sample.
- Likelyhood free inference: allows to evaluate posterior distributions without having to calculate likelihoods.
- Uncertainty quantification: quantifying uncertainties associated with model calculations of true, physical quantities of interest.
Optimal Transport
- Computational OT: define geometric tools that are useful to compare probability distributions.
- Sliced-Wasserstein distance: compares high-dimensional distributions by comparing their projected 1d-distributions.
Works
Pre-print
- F. Coeurdoux, N. Dobigeon, P. Chainais. Plug-and-Play split Gibbs sampler: embedding deep generative priors in Bayesian inference. arXiv pdf code
- F. Coeurdoux, N. Dobigeon, P. Chainais. Normalizing flow sampling with Langevin dynamics in the latent space. arXiv pdf
International Conferences
- F. Coeurdoux, N. Dobigeon, P. Chainais. Learning Optimal Transport Between two Empirical Distributions with Normalizing Flows. ECML-PKDD 2022 arXiv pdf code press
- F. Coeurdoux, N. Dobigeon, P. Chainais. Sliced-Wasserstein normalizing flows: beyond maximum likelihood training. ESANN 2022 arXiv pdf
National conference papers
- F. Coeurdoux, N. Dobigeon, P. Chainais. Approximation du transport optimal entre distributions empiriques par flux de normalisation. Colloque GRETSI, Sept 2022, Nancy pdf
- F. Coeurdoux, N. Dobigeon, P. Chainais. Méthode MCMC plug-and-play avec a priori génératif profond. Colloque GRETSI, Aug 2023, Grenoble pdf
Teaching
2021-2022
- Algorithms and C++ Programming - INP-ENSEEIHT (Toulouse, France)
- Optimization - INP-ENSEEIHT (Toulouse, France)
- Lebesgue integration - INP-ENSEEIHT (Toulouse, France)
- Algorithms and imperative programming - INP-ENSEEIHT (Toulouse, France)
- Statistics - INP-ENSEEIHT (Toulouse, France)
Experience
Laboratory life
Jobs & internships
- Research Engineer — AssessFirst
- Mathematical psychology Designed and developed mathematical and computational model to predict humans behaviours.
- Operationalization Management Streamlining the process of taking machine learning models to production, and then maintaining and monitoring them.
- R&D Engineer - Deep Learning — Credit Mutuel
- Clustering Designed and developed a fast clustering methodology based on DTW distance for large-scale multivariate time series data representing checking account balance.
- Forecasting Designed and developed a probabilistic global forecasting algorithm for millions of time-series with deep learning techniques based on DeepAR.
- Deployement Deployment of a scalable data pipeline for the clustering and forecasting model curently used daily by 1.2 millions users.
- Character Recognition Implementation and deployment of a computer vision algorithms with TensorFlow. Creation of an Optical Character Recognition (OCR) model to authenticate bank contracts.
- Automatization Creation of an NLP tool to automate the loan agreement editing process.
- Software Engineer intern — Beaumanoir Group
- Data Pipeline Automated ETL processes making it easier to wrangle data and reducing time by much as 40%.
- Data Lake Setting up a Microsoft Azure Data Lake, ETLs and data cleanup tasks with Python and XSLT.
- Workflows Updated data streamlining processes, resulting in a 25% redundancy reduction.
Links