Descriptif

Good and expressive data representations can improve the accuracy of machine learning problems and ease interpretability and transfer. For vision tasks, handcrafting good data representations, a.k.a. feature engineering, was traditionally hard. Deep Learning has changed this paradigm by allowing to automatically discover good representations from data. This is known as representation learning. The objective of this course is to provide an introduction to representation learning in computer vision and medical imaging applications. 

We will cover the following subjects: 

- Introduction to Representation Learning for Vision  

- Transfer Learning and Domain Adaptation 

- Self-supervised and Contrastive Learning 

- Knowledge Distillation 

- Disentangled Representations 

- Conditional Generative models 

- Attention and Transformers  

- Visualisation and interpretability in Neural Networks 

- Multimodal representation learning and Foundation models