Courses and Workshop




Courses :

  • Cours 1: Bassem Ben Hamed, Professor at Ecole Nationale d'Electronique et des Télécommunications de Sfax : « Machine Learning and Deep Learning ». This comprehensive course provides a solid grounding in the fields of machine learning and deep learning. It covers the essential concepts, algorithms and techniques used to develop intelligent systems capable of learning from data. Participants will delve into the principles of supervised and unsupervised learning, feature engineering, model evaluation and optimization. In addition, the course explores the exciting field of deep learning, focusing on neural networks and their applications in image recognition, natural language processing, and much more.


  • Cours 2: Angelo Iollo , Professor  at Institut de Mathématiques de Bordeaux: « Introduction to Model Reduction and Projection-Based Techniques ». The course covers model reduction and projection-based techniques in six classes. Topics include an introduction to model reduction and its applications, projection-based reduction methods such as proper orthogonal decomposition (POD) and Galerkin projection, empirical quadratures and their integration, and nonlinear interpolation and mapping techniques. The course concludes with applications to shallow water equations and compressible Euler equations. Practical sessions and numerical examples are integrated to reinforce learning.


  • Cours 3: Enrique ZUAZUA, professor at Friedrich Alexander University. « Control and Machine Learning », In this course, we will present some recent results on the interaction between control and machine learning, and more specifically, supervised learning and universal approximation. We will adopt the perspective of simultaneous or ensemble control of residual neural network systems (ResNets). Roughly speaking, each item to be classified corresponds to a different starting point for the ResNets Cauchy problem, leading to a set of solutions to be steered towards the corresponding targets, associated with the labels, by means of the same control. We present a truly non-linear and constructive method, showing that such an ambitious goal can be achieved, by estimating the complexity of the control strategies. This property is rarely fulfilled by classical dynamical systems in mechanics, and the highly non-linear nature of the activation function governing ResNet dynamics plays a decisive role. It allows half of the phase space to be deformed, while the other half remains invariant, a property not fulfilled by classical mechanical models.
    The turnpike property is also analyzed in this context, showing that an appropriate choice of the cost function used to drive the ResNet leads to more stable and robust dynamics.


  • Cours 4: Sinda Ben Salem, InstaDeep's Product Lead for DeepPack: « Reinforcement Learning for decision-making problems ». The Reinforcement learning (RL) is a paradigm in artificial intelligence that addresses decision-making problems in various domains. RL involves a framework where an agent interacts with an environment to learn optimal strategies through trial and error. The agent aims to maximize reward by adopting adaptive behavior through a series of sequential decisions. RL has found applications in robotics, logistics and games, among others. In this workshop, we will present some of the real-life scenarios in which our research engineers use RL models to improve and streamline our products and solutions.


  • Cours 5: Clement Royer, Professor at Université de Paris Dauphine:  « Optimisation for Machine Learning ». The advent of deep neural networks has given rise to a number of important issues in machine learning, which have a direct impact on the associated optimization problems and methods.
    This lecture aims to present recent developments in this field, focusing on techniques with theoretical foundations applicable to a machine learning context. In this lecture, we will focus on first-order methods, which form the backbone of modern optimization techniques in data science. The first part of the lecture will review gradient descent techniques, presenting recent advances in the application of these methods to convex and non-convex problems.
    The basic gradient descent algorithm and several extensions (proximal, accelerated, on variety,...) will be presented, as well as typical problems of interest. In the second part of the course, we dive into stochastic gradient methods and their relevance to data science problems. Building on the first part, we'll examine the main features of stochastic gradient methods, such as momentum inclusion and mini-lot. We'll also look at popular variants that have been widely used in deep learning applications. Finally, time permitting, we will explore the development of distributed algorithms in the context of large amounts of data, and the algorithmic challenges posed by this context.


  • Cours 6: Hachem Kadri, professor at Université Marseille. « From Classical to Quantum Machine Learning », This course is an introduction to the field of quantum machine learning. The course begins with a general overview of the fundamental concepts and practical applications of classical machine learning, with a particular focus on the most common machine learning algorithms. It then introduces the main concepts of quantum machine learning and gives an overview of the field. To illustrate these concepts with concrete examples, quantum versions of the perceptron and linear regression algorithms are examined.




Training :


  • Training 1: Marwa Hasni and Meriam Ben Youssef, professors at ENIT and Pristini School of AI. « Sequence-to-sequence Deep Learning Models for Time Series Forcasting »


  • Training 2: Ikram Chairi, professor  at Université Mohammed VI Polytechnique. « Sample Selection theory », The four days program is devoted to exploring sample selection biases and techniques for overcoming them in unbalanced data sets. The first day covers sample selection theory, the different biases and an application to the unbalanced distribution. The second day focuses on resampling techniques and a cost-sensitive approach to dealing with biases. The third day is dedicated to the application of cost-sensitive techniques to unbalanced data. The fourth day highlights active and kernel-based learning techniques for these data.




 Workshop Program


Atelier Professionnel : Data Science

Schedule

Saturday 22-02

Monday 24-02

14h30-15h15

 Hatem Zaag

Nejib Zemzemi

15h15-16h00

Lyes Ben Rayana

Safa Layeb Bhar

16h00-16h45

Taoufik Amri

Nozha Boujemaa






Intervenants:


  • Hatem Zaag (Université de Sorbonne Paris Nord, France). Une description détaillée de la conférence sera ajoutée ultérieurement.
  • Lyes Ben Rayana (BIAT ,Value, Tunisia).  Une description détaillée de la conférence sera ajoutée ultérieurement.
  • Taoufik Amri, Senior data Scientist, Cap Gemini, France). Une description détaillée de la conférence sera ajoutée ultérieurement.
  • Nejib Zemzemi (Research scientist INRIA Bordeaux Sud-Ouest). Une description détaillée de la conférence sera ajoutée ultérieurement.
  • Safa Layeb Bhar (ENIT). Une description détaillée de la conférence sera ajoutée ultérieurement.
  • Nozha Boujemaa (Decathlon, France). Une description détaillée de la conférence sera ajoutée ultérieurement.