personal page
In a recent interview with Alexandre Boero from Clubic we discuss how recent technologies rendered possible a growing network of fake online media sites and journalists entirely generated by AI, designed to appear credible and manipulate audiences and advertisers, raising serious concerns about misinformation and the erosion of trust in digital content.
Deepfakes, AI-generated articles, fake websites mimicking the media… in a recent interview with Marie-Eve Frénay from Les Echos and we discuss the new generation of manipulative content flooding the web.
This joint work with Laetitia LAGUZET has been presented at the 5th Computational Methods for Multi-scale, Multi-uncertainty and Multi-physics Problems Conference held in Porto, 1-4 July 2025.
Abstract Physics-Informed Neural Networks (PINNs) are a type of neural network designed to incorporate physical laws directly into their learning process. These networks can model and predict solutions for complex physical systems, even with limited or incomplete data, often using a mathematical formulation of a state equation supplemented with other information.
Introduced by Raissi et al. (2019), PINNs find applications in fields like physics, engineering, and fluid mechanics, particularly for solving partial differential equations (PDEs) and other dynamic systems. In this contribution we explore a modification of PINNs to multi-physics numerical simulation involving radiation transport equations; these equations describe the propagation of a Marshak-type wave in a temperature dependent opaque medium and is considered a good benchmark for difficult multi-regime computations.
This joint work with Pierre Brugière has been presented at the at the 8th Future of Information and Communication Conference 2025 held in Berlin, 28-29 April 2025.
Talk materials:
Abstract : The transformer models have been extensively used with good results in a wide area of machine learning applications including Large Language Models and image generation. Here, we inquire on the applicability of this approach to financial time series. We first describe the dataset construction for two prototypical situations: a mean reverting synthetic Ornstein-Uhlenbeck process on one hand and real S&P500 data on the other hand. Then, we present in detail the proposed Transformer architecture and finally we discuss some encouraging results. For the synthetic data we predict rather accuratly the next move, and for the S&P500 we get some interesting results related to quadratic variation and volatility prediction.
Teacher: Gabriel TURINICI
Content
Historical note: 2019/21 course name: « Approches déterministes et stochastiques pour la valuation d’options » + .
Instructor: Gabriel TURINICI
Preamble: this course is just but an introduction, in a limited amount of time, to Statistical and Machine learning. This will prepare for the next year’s courses (some of them on my www page cf. « Deep Learning » and « Reinforcement Learning »).
1/ Introduction to statistical learning : supervised, non-supervised and reinforcement learning, general learning procedure, model evaluation, under and overfitting
2/ K-nearest neighbors and the « curse of the dimensionality »
3/ Regression in high dimensions, variable selection and model regularization (ridge, lasso)
4/ Stochastic gradient descent, mini-batch
5/ Neural networks: introduction, operator, datasets, training, examples, implementations
6/ K-means clustering
Main document for the theoretical presentations: (no distribution autoried without WRITTEN consent from the author): see your « teams » group.
Exercices, implementations: see « teams » group.
Responsable de cours: Gabriel TURINICI
Contenu:
1 Introduction
2 EDO
3 Calcul de dérivée et contrôle
4 EDS
Bibliographie: poly distribué
Supports de cours: livre en anglais « Numerical simulations of time-dependent problems : applied to epidemiology, artificial intelligence and finance »
POLY en français (mis à jour 11/03/2024). Ne pas diffuser sauf accord ÉCRIT de l’auteur. Attention: seulement les cours en amphi font foi, le poly est juste pour aider.Implementations TP:
EDO: exo sur la précision, exo stabilité , SIR(EE+H+RK4), order for the EE/H schemes/SIR
SIR (version controle, adjoint / backward); (version 2023 here)
EDS version 2025 : implémenter :
1/ simulation brownien
2/ calcul intégrale de W par/rapport à W
3/ calcul par Euler-Maruyama pour équation d’Ornstein-Uhlenbeck
4/ calcul par Euler-Maruyama faible pour modèle log-normal (Black-Scholes)
2021/22 ici)
Cours du 07/03/2023 (version confinement 2021!!): video Youtube chapitre 3 (motivation calcul backward) : 
visioner ICI
livre « Simulations numériques des problèmes dépendant du temps: appliquées à l’épidémiologie, l’intelligence artificielle et les finances« ici. prises lors des séances de cours.séance de TD (groupe 1): séance 1: ICI, séance 2, séance 3, séance 4, séance 5, séance 6,séance 7 (EDS).Cours 1 du 19/1/2021: video Youtube chapitre 1 (motivation/EDO) : visioner ICI , ensuite revenir sur la séance Teams pour commentaires et questions.; video Youtube chapitre 1 (motivation/EDS) : visioner ICI , ensuite revenir sur la séance Teams pour commentaires et questions.; video Youtube chapitre 1 (motivation/backward) : visioner ICI , ensuite revenir sur la séance Teams pour commentaires et questions.Cours 2: voir le poly annoté; une video Youtube partielle chapitre 2 (EDO/existence) : visioner ICIThis joint work with Stefana-Lucia ANITA has been presented at the at the 27th International Conference on Pattern Recognition (ICPR) 2024 held in Kolkata, India, Dec 1st through 5th 2024.
Talk materials:
Abstract : Although Multi Armed Bandit (MAB) on one hand and the policy gradient approach on the other hand are among the most used frameworks of Reinforcement Learning, the theoretical properties of the policy gradient algorithm used for MAB have not been given enough attention. We investigate in this work the convergence of such a procedure for the situation when a L2 regularization term is present jointly with the ‘softmax’ parametrization. We prove convergence under appropriate technical hypotheses and test numerically the procedure including situations beyond the theoretical setting. The tests show that a time dependent regularized procedure can improve over the canonical approach especially when the initial guess is far from the solution.
This talk has been presented at the at the 27th International Conference on Pattern Recognition (ICPR) 2024 held in Kolkata, India, Dec 1st through 5th 2024.
Talk materials:
Abtract : Physics-informed neural networks (PINN) is a extremely powerful paradigm used to solve equations encountered in scientific computing applications. An important part of the procedure is the minimization of the equation residual which includes, when the equation is time-dependent, a time sampling. It was argued in the literature that the sampling need not be uniform but should overweight initial time instants, but no rigorous explanation was provided for this choice. In the present work we take some prototypical examples and, under standard hypothesis concerning the neural network convergence, we show that the optimal time sampling follows a (truncated) exponential distribution. In particular we explain when is best to use uniform time sampling and when one should not. The findings are illustrated with numerical examples on linear equation, Burgers’ equation and the Lorenz system.
Teacher: Gabriel TURINICI
Summary:
1/ Deep learning : major applications, references, culture
2/ Types: supervised, renforcement, unsupervised
3/ Neural networks: main objects: neurons, operations, loss fonction, optimization, architecture
4/ Stochastic optimization algorithms and convergence proof for SGD
5/ Gradient computation by « back-propagation »
6/ Pure Python implementation of a fully connected sequential network
7/ Convolutional networks (CNN) : filters, layers, architectures.
8/ Keras implementation of a CNN.
9/ Techniques: regularization, hyper-parameters, particular networks, recurrent (RNN, LSTM);
10/ Unsupervised Deep learning: generative AI, GAN, VAE, Stable diffusion.
11/ Keras VAE implementation. “Hugginface” Stable Diffusion.
(12/ If time allows: LLM & NLP: word2vec, Glove (exemples : woman-man + king = queen)
| Documents | ||
| MAIN document (theory): see your teams channel (no distribution is authorized without WRITTEN consent from the author) | for back-propagation | SGD convergence proof |
| Implementations |
| Function approximation by NN : notebook version, Python version Results (approximation & convergence)     After 5 times more epochs   Official code reference https://doi.org/10.5281/zenodo.7220367 |
| Pure python (no keras, no tensorflow, no Pytorch) implementation (cf. also theoretical doc): – version « to implement » (with Dense/FC layers) (bd=iris), – version : solution If needed: iris dataset here Implementation : keras/Iris CNN example: https://www.tensorflow.org/tutorials/images/cnn Todo : use on MNIST, try to obtain high accuracy on MNIST, CIFAR10. |
| VAE: latent space visualisation : CVAE – python (rename *.py) , CVAE ipynb version |
| Stable diffusion: Working example jan 2025: python version, Notebook version Old working example 19/1/2024 on Google collab: version : notebook, (here python, rename *.py). ATTENTION the run takes 10 minutes (first time) then is somehow faster (just change the prompt text).  |
