Deep

Deep learning

Description: This will cover all the components of deep learning, a sub-field of machine learning that has seen enormous progress in recent years. In particular, we will discuss architectures (multi-layer, convolutional and recurrent networks), computation graphs and optimisation of the parameters of these computation graphs, and regularisation techniques. The ‘in-depth’ course will focus on computer vision. In particular, various architectures emerging in recent years will be detailed, in particular Vision Transformers, adversarial generative and diffusion models as well as self-supervised learning approaches. The implementation of the concepts discussed in class will be carried out in pytorch.

Assessment through a challenge immerses students in the context of AI research, where scientific results are frequently presented and compared in this way.

Evaluation methods: Taking part in a challenge, cannot be retaken.

Evaluated skills:

• Modelling
• Research and Development
• Management

Course supervisor: Jérémy Fix

Geode ID: 3MD4040

External resources:

• Site du cours
• Sujets des TPs