Duration
30h Th, 60h Proj.
Number of credits
Lecturer
Language(s) of instruction
English language
Organisation and examination
Teaching in the second semester
Schedule
Units courses prerequisite and corequisite
Prerequisite or corequisite units are presented within each program
Learning unit contents
Deep learning underpins most of what currently works in computer vision, language processing and scientific modelling. This course is about how those models work, and the only way to find out is to build them. The mathematics is derived at the blackboard, step by step: backpropagation, training objectives, variational bounds, diffusion formulations. Each lecture comes with code in which the algorithms are implemented and run.
The course is demanding. Lectures are dense, mathematical and fast. In exchange, you come out able to read a current paper, implement it, and understand why it works.
Topics covered:
- Fundamentals of machine learning
- Multi-layer perceptron
- Automatic differentiation
- Training neural networks
- Convolutional neural networks
- Computer vision
- Attention and transformers
- GPT, from scratch
- Graph neural networks
- Uncertainty
- Auto-encoders and variational auto-encoders
- Diffusion models
Learning outcomes of the learning unit
At the end of the course, you will be able to derive the core algorithms of deep learning rather than recite them, train a neural network while understanding what every layer of the stack does, diagnose why a model fails to learn, and reproduce a recent paper. You will also know what remains unsolved.
The project is where this comes together. You choose a problem, build a model for it over the semester, and present the result at a poster session.
Prerequisite knowledge and skills
Solid programming experience, in Python in particular. Probability, statistics and linear algebra, at a level where you can use them, not just recognise them.
Taking "ELEN0062 Introduction to Machine learning" first is **strongly** recommended. Students who skip it usually spend the first weeks catching up on material the lectures assume.
Planned learning activities and teaching methods
- Theoretical lectures
- Programming project, presented at a poster session
Mode of delivery (face to face, distance learning, hybrid learning)
Face-to-face course
Further information:
Lectures will taught face-to-face. The project is carried out remotely.
Course materials and recommended or required readings
Other site(s) used for course materials
- Github (github.com/glouppe/info8010-deep-learning)
Further information:
Slides are made publicly available on GitHub during the semester.
Exam(s) in session
Any session
- In-person
oral exam
Written work / report
Further information:
The evaluation has two components:
- Programming project: 40%
- Oral exam on the theory: 60%
The project is presented at a poster session at the end of the semester, where you give an overview of what you did, motivate your design choices and discuss your results.
Questions on the code are asked separately, right after the oral exam on the theory. You answer direct questions on your own implementation and may be asked to modify or extend it. This counts towards the project grade.
A minimum of 8/20 is required for each component. Below that, the lowest of the two grades becomes the final grade.
Use of AI tools
Using AI assistants is allowed, without restriction. The rule is simple: everything you submit, you must be able to explain. Submitting a solution you do not understand, because an assistant produced it, will be strongly penalized. This is why the evaluation is built the way it is: the exam is oral, the poster session is oral, and the questions on your code are asked with you in the room. A model can write your project, but it cannot sit the exam, explain why you picked that architecture, or modify your implementation on request.
Work placement(s)
Organisational remarks and main changes to the course
The website for the course is https://github.com/glouppe/info8010-deep-learning
Contacts
- Teacher: Prof. Gilles Louppe (g.louppe@uliege.be)