Patrick Saux

Patrick Saux

PhD in Reinforcement Learning

Inria Scool

CRIStAL (CNRS)

Biography

I obtained my PhD in mathematics (theoretical statistics) & computer science (reinforcement learning) at Inria Lille (team Scool, formerly SequeL) & CRIStAL (CNRS) in January 2024, under the supervision of Philippe Preux and Odalric-Ambrym Maillard.

My research focused on reinforcement learning and stochastic bandits in risky, nonparametric environments. I also collaborated with François Pattou (PU-PH at CHU Lille and Inserm) and his team to design personalised, data-driven follow-up programmes after bariatric surgery for patients living with obesity.

Interests
  • Stochastic Processes
  • Reinforcement Learning
  • Stochastic Bandits
  • Nonasymptotic statistics
  • AI for Health
  • Obesity and Bariatric Surgery
Education
  • PhD in Mathematics/Computer Science, 2024

    Inria Scool.

  • MSc in Machine Learning (MVA), 2020

    École Normale Supérieure Paris-Saclay

  • MSc in Probability and Finance (El Karoui), 2016

    École Polytechnique and Sorbonne Université (Paris VI)

  • Bs in Mathematics (M1), 2015

    École Normale Supérieure Paris-Saclay

  • École Préparatoire MPSI/MP*, 2013

    Lycée Privé Sainte-Geneviève

Research

Experience

 
 
 
 
 
Teaching Assistant
Jan 2022 – Mar 2022 Palaiseau
Reinforcement learning class (MAP/INF641).
 
 
 
 
 
Teaching Assistant
Mar 2021 – Apr 2021 Gif-sur-Yvette
Reinforcement learning class.
 
 
 
 
 
Data Scientist
Jun 2020 – Oct 2020 Paris
Quantitative trading strategies on crude oil based on estimated storage from satellite imaging.
 
 
 
 
 
Associate Quant Trading Strategist
Goldman Sachs
Apr 2016 – Sep 2019 London

Interest rates volatility, flow credit.

  • Modelling credit defaults and rates volatility using stochastic processes and Monte Carlo simulation.
  • Implementation using Slang and SecDb proprietary tools, monitoring of real-time, distributed systems.
  • Bond trading strategies (corporate, illiquids, structured notes, callables).
 
 
 
 
 
Research Intern
Mar 2015 – Aug 2015 Oxford

Oxford-Man Institute is a research lab focused on mathematics, machine learning and quantitative finance, jointly led by Oxford University and Man AHL. My research was supervised by Terry Lyons.

  • Rough paths theory, an abstract analytical and algebraic framework to study complex and noisy signals.
  • Application of path signature to data analysis.
  • Prediction of seismic events on real-world mining data.

Academic duties

Conference Reviewer

AISTATS 2022 (top reviewer)

Workshop Reviewer

EWRL 2022

Conference Reviewer

AISTATS 2023 (top reviewer)

Projects

Bariatric Weight Trajectory Prediction

Bariatric Weight Trajectory Prediction

An online prediction tool for weight loss trajectory after bariatric surgery.

Stochastic

Stochastic

All sorts of stochastic simulations.

ito-diffusions

ito-diffusions

Libraries for stochastic processes simulation and visualization.

Bond pricer

Bond pricer

Interactive bond pricer, yield calculation and Monte Carlo pricing for callable.

Expander graphs

Expander graphs

Implementation of several expanders and empirical evidence of spectral and connectivity properties. Contribution to the networkx library.

Random matrices

Random matrices

Web app on spectral properties of large random matrices.

Markov Epidemic

Markov Epidemic

Markov stochastic models (SIS, SIR, SEIR…) to describe the evolution of epidemics on a network of connected individuals.

Neural exploration

Neural exploration

Analysis and implementation of neural approximator for contextual bandits and episodic MDP.

Continuity of Graph Embeddings

Continuity of Graph Embeddings

Theoretical and practical continuity of various graph embeddings (eigenmap, deep walk, graph kernels, random walk factorisation…)

Optimal Transport Correlation

Optimal Transport Correlation

Geometric study of correlation matrix via Frechet mean.

Connectivity Loss

Connectivity Loss

Train robust autoencoder with topological loss.

Fractal

Fractal

Fractal generation in Python using Just-In-Time (JIT) compilation and Numba.

rlberry

rlberry

A Reinforcement Learning Library for Research and Education.

Contact

  • firstname[dot]jr[dot]surname[at]gmail[dot]com