Descriptif
MAP670R: Advanced topics in Deep Learning
(2023)
Lecturer: Edouard Oyallon
Description :
The objective of this class is to discuss several existing theoretical and practical
results about deep neural networks. Those later consist in a cascade of linear and
non-linear pointwise operators that are typically used for regression or generative
tasks. They are now standard in many machine learning applications because
they lead to outstanding performances. However, they are poorly understood
from a theoretical perspective and they require many ad-hoc engineering tricks
to be successfully trained. This class proposes to discuss several recent results,
in simplified settings, through the lens of signal processing tools.
Format des notes
Numérique sur 20Littérale/grade réduitPour les étudiants du diplôme M2 Data Sciences
Le rattrapage est autorisé (Max entre les deux notes)- Crédits ECTS acquis : 3 ECTS
Programme détaillé
Material A manuscript linked to the class can be found here: https://edouardoyallon.github.io/ATDL-
2022.pdf as well as last year webpage: https://edouardoyallon.github.io/MAP670R-
2021/index.html
