Biography

Pierre Bras is a third-year PhD student at LPSM Sorbonne Université under the direction of Gilles Pagès. His research focuses on numerical methods for probability and statistics with applications to Machine Learning and Finance.

Interests
  • Machine Learning
  • Numerical Probability
  • Stochastic Optimization
  • Monte Carlo methods
  • Mathematical Finance
  • Stochastic control and Reinforcement Learning
Education

  • PhD in Applied Mathematics, 2020-2023

    Sorbonne Université, LPSM

  • Master 2 Probability and Random Models, option Applied Probability, 2019

    Sorbonne Université, LPSM

  • Diploma of Ecole Normale Supérieure in Mathematics and Applications, 2016-2020

    Ecole Normale Supérieure de la rue d'Ulm

Publications & pre-Publications

Pierre Bras, Gilles Pagès (2023). Reinforcement Learning for Variance Reduction in Monte Carlo simulation. In progress.

Cite

Pierre Bras, Gilles Pagès (2022). Langevin algorithms for Markovian Neural Networks and Deep Stochastic control. In arXiv e-prints.

PDF Cite arXiv GitHub

Pierre Bras (2022). A note on $L^1$-Convergence of the Empiric Minimizer for unbounded functions with fast growth. In arXiv e-prints.

PDF Cite arXiv

Pierre Bras (2022). Langevin algorithms for very deep Neural Networks with applications to image classification. In arXiv e-prints.

PDF Cite arXiv GitHub

See all publications

Recent & Upcoming Talks

14th International Conference on Monte Carlo Methods and Applications
Convergence of Langevin-Simulated Annealing algorithms with multiplicative noise
Jun 28, 2023 12:00 PM Sorbonne Université Campus Pierre et Marie Curie
International Neural Network Society Workshop on Deep Learning Innovations and Applications INNS DLIA 2023
Langevin Algorithms for Very Deep Neural Networks with application to Image Classification
Jun 23, 2023 Gold Coast Convention and Exhibition Centre, Australia
International Joint Conference on Neural Networks IJCNN 2023
Langevin Algorithms for Markovian Neural Networks and Deep Stochastic Control
Jun 18, 2023 — Jun 23, 2023 Gold Coast Convention and Exhibition Centre, Australia
See all events

Code & Jupyter Notebooks

Langevin algorithms for Markovian Neural Networks and Deep Stochastic control
In this repository we give the implementation of Langevin and Layer Langevin optimizers as instances of the TensorFlow tf.keras.optimizers.Optimizer base class and we compare Langevin and non-Langevin optimizers for the training of various stochastic control problems.
Langevin algorithms for Markovian Neural Networks and Deep Stochastic control
Langevin algorithms for very deep Neural Networks with applications to image classification
In this repository we give the implementation of Langevin and Layer Langevin optimizers as instances of the TensorFlow tf.keras.optimizers.Optimizer base class and we compare Langevin and non-Langevin optimizers for the training of various problems for image classification.
Langevin algorithms for very deep Neural Networks with applications to image classification
Convergence rates of Gibbs measures with degenerate minimum
This repository contains a notebook showing how we can compute the higher order nested expansion of some objective function to be minimized in problems where the Hessian matrix at the minimum is singular. In this example, we train a neural network and study the behaviour of the loss function in the neighbourhood of the (empirical) minimum. The machine learning library that is used is TensorFlow.
Convergence rates of Gibbs measures with degenerate minimum
See all

Contact