Skip to content

AI Fundamentals

PDF icon Download in PDF

  • Lecturer: Daniel Racoceanu (daniel.racoceanu@sorbonne-universite.fr)
  • Course code: UM4RBI10-FondIA
  • Student workload: 14h of lectures, h of tutorials, 14h of labs
  • Credits: 3 ECTS
  • Specialization tracks:
  • Semester offered: S1 S2 S3 S4
  • Language of instruction: French English
  • Targeted audience: Eng. Sc. department Other :
  • Localization : PMC Campus Other :

Course Overview

This course offers a deepening of the basic concepts used in artificial intelligence and machine learning, with a focus on fundamental paradigms/algorithms and theoretical and methodological reinforcement

Mots-clés : Supervised, unsupervised, self-supervised learning; Markov processes, Bayesian approaches, frequentist approaches, inverse problems, decision trees, strong randomness, ensemble learning, support vector machines, kernel trick, loss function, evaluation methodologies, bias, stochastic optimization, gluttonous approaches.

Prerequisites

Students should have previously acquired the following prerequisites to follow this course:

  • Mathematics: Linear algebra, vector and matrix calculus, numerical optimization of multivariate continuous functions, probability
  • Computer science: Programming (in Python: numpy, etc.), algorithms
  • Machine Learning (desirable): Machine Learning Basics, Machine Learning Performance Assessment (False Positive, False Negatives, Recall, Accuracy, F1-score)

Intended Learning Outcomes

By the end of this course, students will be able to:

  1. Analyze a problem: Choose an algorithm that is appropriate for the task, the data, and the resources
  2. Design: Algorithmic Foundations
  3. Design: Choosing the learning process
  4. Implementation: Master a high-level library dedicated to learning to implement the algorithm
  5. Implementation: Preparing data for training
  6. Implementation: Perform the training defined in the design phase to optimize the algorithm's parameters and meta parameters
  7. Implementation: Replicating learnings and experimental results
  8. Assessment: Develop an adequate experimental protocol to evaluate the performance of a model
  9. Assessment: Diagnose the causes of suboptimal network behavior in order to propose solutions to address it
  10. Assessment: Interpreting and explaining the decisions of a model: attention map, bias checking, etc.

Indicative Teaching Sequence and Methods

Week C/TD/TP* Content Preparation Learn.\ outc.
S1 C1 Markov processes, dynamic programming, Bellman's equations, link with reinforcement learning Markov property, Markov processes, dynamic programming, reinforcement learning. AAV1,2,3
S2 TP1 (3h) Reinforcement learning: use of Markov processes and dynamic programming in this context. Typical use cases. Look in detail at the concepts covered in C1 AAV1-6,8
S3 C2 Bayesian vs. frequentist approaches, Bayesian networks, link with inverse problems. Bayes' theorem, definition of inverse problems. AAV1,2,3
S4 TP2 (3h) Bayesian systems and Bayesian approaches vs. frequentist approaches, inverse problems Look in detail at the concepts covered in C2 AAV1-6,8
S5 C3 Stochastic optimization in discrete space AAV1,2,3
S6 TP3 (3h) Stochastic optimization: discrete and continuous spaces Look in detail at the concepts covered in courses C1 to C3 AAV1-6,8
S7 C4 Stochastic optimization in \(R^n\) discrete space, continuous space AAV1,2,3
S8 TP4 (3h) Stochastic optimization and reinforcement learning Take a detailed look at the concepts covered in course C1 to C4 AAV1-6,8
S9 C5 Decision tree, random drills, set learning Decision tree, random forests AAV1,2,3
S10 C6 Support Vector Machines (SVMs): Fundamentally statistical nature of the approach, kernel trick; SVM - kernel tip AAV1,2,3
S11 C7 Methodologies for evaluation/validation and bias checking. Evaluation, verification, bias AAV 5,6,7,8, 10
S12 TP5 (2h) Validation and bias verification methodologies. Use cases inspired by recent advances in the field. Look in detail at the concepts covered in C7 AAV 5,6,7,8,10
  • C/TD/TP respectively corresponds to lectures, tutorials and lab sessions.

Sequence of the unit

The UE takes place in 3 successive sequences:

  • Sequence 1: Markov processes / Bayesian approaches (C1/2 + TP1/2 + Quiz1)

  • Sequence 2: stochastic optimization (C4/5 + TP3/4 + Quiz 2)

  • Sequence 3: convergence / loss function + evaluation - bias (C5/6/7 + TP5) + overall ER (all courses and practical work)

The practical exercises will have a TME part (on a machine) at the beginning of the session, in order to be able to instantaneous the concepts of the course, in an easy way, before the actual start of the practical exercise.

Indicative Assessment of Intended Learning Outcomes (1st session)

Week Individ./group In-person/remote Type of exam Evaluated outcomes Scale %
S6 Collective In-person Quiz AAV1-6,8 25%
S8 Collective In-person Quiz AAV1-6,8 12%
S10 Collective In-person Quiz AAV1-6,8 12%
S15 Collective In-person Written AAV 1-10 50%

2nde session

Session Individ./group In-person/remote Type of exam Evaluated outcomes Scale %
2 Collective In-person Written AAV 1-10 50%
1 Collective In-person Quiz AAV1-8 50%

Bibliographic references

  • Stuart Russell \& Peter Norvig, Artificial Intelligence : A Modern Approach, Fourth edition, 2020
  • Richard S. Sutton et Andrew G. Barto, Reinforcement Learning : An Introduction, second edition , 2018

Logo SDI Date of generation of this unit description: 14/01/2026 Logo SDI