The third year is first taught in September
2027, to the cohort that entered in
2025. What follows is the current draft of the year: courses,
credits and teaching assignments are still being settled and
will change.
The third year specializes. A last layer of foundations – measure
and integration, the analysis of algorithms, advanced databases –
carries courses on the fields artificial intelligence is applied to:
language processing, computer vision, generative models and
conversational agents, robotics, and the mechanisms that govern how
agents interact. A data challenge runs across the first semester,
and trustworthy AI takes up in technical terms what the ethics and
law of the first two years raise.
The second semester is a short, intensive block of teaching
followed by three to four months in a laboratory or a company. Every
course is taken by everyone: the choice this year asks for is
between a research internship and an industrial placement, and the
subject of the work that goes with it.
Semester 5 · September to January
7 courses ·
30 ECTS
· 366 contact hours
Foundations of AI
Measure, Integration and Probability7 ECTS
Mathematics84 h14 weeks3 h lectures per week3 h tutorials per week
Measures and the Lebesgue integral, the convergence theorems,
product measures, and the spaces of integrable functions – with
the Hilbert space structure of the square-integrable ones,
orthonormal bases, and Fourier series.
Then probability rebuilt on that ground: random variables as
measurable functions, independence, conditional expectation,
characteristic functions, the modes of convergence, and proofs
of the law of large numbers and the central limit theorem. It
is the language conditional expectation, Bayesian inference and
Markov chains are written in.
Advanced Analysis of Algorithms7 ECTS
Computer science84 h14 weeks3 h lectures per week3 h tutorials per week
Analysing algorithms rather than only writing them: amortized
and average-case analysis, randomization and what it buys,
approximation algorithms and the limits of approximation,
competitive analysis of algorithms that must answer before they
have seen everything, and algorithms for data too large to hold
in memory. Takes up where the computability and NP-hardness of
year 2 left off, on the question of what can be computed
well.
Advanced Databases3 ECTS
Computer science42 h14 weeks1.5 h lectures per week1.5 h tutorials per week
What a database engine does inside: query evaluation and
optimization, transactions and concurrency, then distributed
storage, streaming data, and data models that are not relational.
Continues the first-year course.
AI-focused courses
Natural Language Processing3 ECTS
Artificial intelligence42 h14 weeks1.5 h lectures per week1.5 h tutorials per week
How language is represented and processed: tokenization and
embeddings, sequence models, attention and transformers, and the
tasks the field is measured on – parsing, translation, question
answering.
Computer Vision3 ECTS
Artificial intelligence42 h14 weeks1.5 h lectures per week1.5 h tutorials per week
Image formation and representation, classification, detection,
and segmentation, and the geometry of several views. Practical
work on real images throughout.
Data Challenge5 ECTS
Machine learning42 h14 weeks3 h lectures per week
Teams take on open modelling problems and work on them across
the semester: public competitions, with a dataset, a metric, and
a leaderboard, or problems brought in by partners. What is
assessed is not only the score: the report and the defence ask
how the data was understood, what was tried, what failed, and
what the model would be worth outside the challenge.
Transverse courses
PSL Week2 ECTS
Transverse30 h1 week
A week of intensive teaching taken outside the programme, chosen
from the courses Université PSL opens to students of all its
member schools.
Semester 6 · teaching February to March, placement from April
6 courses ·
30 ECTS
· 180 contact hours in the timetabled courses
AI-focused courses
Generative AI and Conversational Agents3 ECTS
Artificial intelligence36 h8 weeks3 h lectures per week1.5 h tutorials per week
Generative models and the systems built on them: how they are
trained, adapted, and prompted, how they are evaluated, how they
are put behind an interface, and how they fail.
Robotics and Autonomous Systems3 ECTS
Artificial intelligence36 h8 weeks3 h lectures per week1.5 h tutorials per week
Perception, control, and planning in systems that act in the
physical world, and what changes when the environment answers
back: uncertainty, real time, and the cost of being wrong.
Trustworthy AI3 ECTS
Machine learning36 h8 weeks3 h lectures per week1.5 h tutorials per week
The technical side of what the ethics and law courses of the
first two years raise: robustness and adversarial examples,
privacy and what can be inferred from a model, explanation
methods and how much they really explain, and the evaluation of
fairness.
Mechanism Design and Game Theory3 ECTS
Computer science36 h8 weeks3 h lectures per week1.5 h tutorials per week
Strategic interaction between agents: equilibria, auctions,
matching and voting, and the design of mechanisms whose outcome
is good even when everyone acts in their own interest.
Topics in AI3 ECTS
Artificial intelligence36 h8 weeks3 h lectures per week1.5 h tutorials per week
A course whose subject changes from year to year, given by a
researcher on a current question in artificial intelligence or
its mathematical foundations.
Research Internship or Industrial Placement15 ECTS
Internship
Three to four months in a research laboratory or in a company,
from April, on a subject agreed with a supervisor. It ends in a
written report and an oral defence, and is where the year’s
specialization actually takes shape.
Course descriptions are written and maintained by
the teaching team. Contact hours are hours of class per student;
credits follow the European ECTS scale, 60 of them per year.
