Machine Learning for NLP: the Fundamentals (Master S1 CLass)

Credits: 6
Format: 4 hours weekly
2026-2027 instructors: Marie Candito
2026-2027 schedule: Tuesdays 09:00-11:00 and 11:15-13:15, room 234C (Halle aux farines)
Moodle page

The course provides the fundamental concepts of supervised classification via deep learning methods, with typical examples in NLP. It will cover:

linear and log-linear models, linear separability and kernel methods
multi-layer perceptrons, non-linearities, universal approximation theorem
learning as loss minimization, stochastic gradient descent and variants, backpropagation algorithm
word embeddings as dense features, word embeddings’ learning
overview of metrics and methodology in supervised machine learning
Depending on progress, we will touch upon how these concepts are used within large language models, thanks to self-supervised learning and the language modeling training objective.

Programming skills in python. Basic linear algebra (dot product, matrix product), basic probability theory (chain rule of probability).

On successful completion of this course, students should be able to:

master the fundamental concepts of machine learning;
understand and design supervised classifiers of various types;
master operations on tensors;
get basic proficiency in some of the most vastly used python machine learning libraries (sklearn and pytorch);
get a first broad picture of how supervised learning is applied within (large) language models, thanks to the notion of self-supervision and the language modeling training objective.