Frugal AI: from mitigating data scarcity to designing functional spaces

We are pleased to invite you to an upcoming seminar presented by Sylvain Chevallier, Full Professor Univ. Paris-Saclay/Inria, 

next Thursday 20 March at 11:00am in BR-B02-007A  (Petit Amphi).

Frugal AI: from mitigating data scarcity to designing functional spaces

Applicative domains with limited data availability are a major challenge in machine learning. Starting with data-oriented approaches, I will introduce methods proposed in Riemannian geometry to consider time series datasets in a well-defined representative space, with applications mainly focusing on brain signal processing. In the second part of the talk, I will introduce a novel approach for growing neural networks by finding the expressivity bottleneck and finding the right architecture to fix it, with applications on computer vision.