Objectifs :
In this course, we aim to closely study modern artificial intelligence methods to better grasp their real potential and understand the ethical and societal issues arising from this technology.

After a brief historical overview of AI development, we will examine recent methodologies for training neural networks using (very) large datasets. We will explore the mathematical fundamentions of such approaches, in particular Gradient Descent and Backpropagation algorithms. We will build operational models to perform basic approximation or classification tasks, with a particular focus on computer vision. Finally, in the last part of the course, we will introduce recent applications of generative AI. We will cover the transformer architecture, which underpins models like ChatGPT and Claude.

In general, this is a descriptive course that primarily introduces concepts. Advanced mathematical formalism is not required, but familiarity with topics taught in the first two years of a scientific undergraduate degree (see a list of prerequisites below) will be helpful. Additionally, as mentioned above, we will conduct several guided practical programming experiments to construct models that illustrate the methods discussed. We will use the Python programming language and its libraries.

Programme :
- Statistical learning and neural networks: from theory to initial practical implementations
- Mathematical foundations of neural networks: gradient descent, backpropagation
- Convolutional neural networks (CNN): image recognition, fine-tuning, transfer learning
- Autoencoders, variational autoencoders
- Transformers (ChatGPT, Bard, etc.)

Pré-requis :
- General mathematics from the first two years of a scientific undergraduate program (matrix calculus, partial derivatives, gradients)
- Basic knowledge of probability and statistics
- Python programming will be used for practical experiments; a beginner level should be sufficient

Equipe pédagogique :
STOS Andrzej

La formation participe à l'objectif suivant :conforter la culture scientifique des doctorants dans leur champ disciplinaire ou en interdisciplinaire