Representation learning is a pillar of modern artificial intelligence. As the volume of data grows exponentially, the capacity of extracting usefull and significant representation from the data is crucial for AI systems performances. Representation learning aims at finding and learning underlying representations in the data. These representations are then used to characterize and exploit the data efficiently and in an informative manner. We develop methods based on auto-supervised learning, generative and adversarial models to learn the best representations.
Related Projects
APATE
A Prototype deepfake Assessment Toolbox for forensic Experts
FIRST
Early Detection of Hematological Abnormalities Using Artificial Intelligence
Related Publications
[1]
Théo Lepage • Réda Dehak. "Label-Efficient Self-Supervised Speaker Verification With Information Maximization and Contrastive Learning". Proceedings of the 23rd Annual Conference of the International Speech Communication Association (Interspeech 2022). 2022. https://doi.org/10.21437/Interspeech.2022-802.