Skip to content
Loading Events

« All Events

  • This event has passed.

AI: Cutting-Edge Overviews & Tutorials

Headshot image of Dr Felipe Tobar

You are invited to join I-X on-site or online via Teams for Dr Felipe Tobar’s Cutting-Edge Overviews & Tutorial Seminar, Optimal Transport for Machine Learning and Data Analysis. 

If you wish to attend in person, please register here for building access purposes. Registration will close on Sunday 2 February at 12:00 GMT.

Abstract:

Optimal transport (OT) provides a general-purpose framework to quantify the discrepancy between two probability distributions by “lifting” a distance defined on their support to the space of probability distributions. In the last decade, the impact of OT in machine learning (ML) cannot be overstated: current OT-powered extensions have been applied to GANs, VAEs, dimensionality reduction, and model selection with applications on genomics, health, finance, audio, robotics, and astrophysics. In this tutorial, I will introduce OT and show its use in the development of ML applications, all with the aim of encouraging the adoption of the OT toolbox by those using AI/ML tools in their scientific research. The tutorial will start with the usual formulations of OT and historical context, to then discuss metric properties, computational considerations, and the celebrated Wasserstein barycentre. We will see demonstrations of the main OT algorithms in the form of Jupyter Notebooks.

Bio:
Felipe Tobar is a Senior Lecturer in Machine Learning at the Department of Mathematics and I-X at Imperial. Previously, he was an Associate Professor at Universidad de Chile (2021-2024), a Researcher at the Center for Mathematical Modeling (2016-2020), a postdoc at the Machine Learning Group, University of Cambridge (2015), and obtained his PhD in Signal Processing from Imperial in 2014. Felipe’s research lies in the intersection of Statistical Machine Learning and Signal Processing, including approximate inference, spectral estimation, optimal transport, diffusion models, and Gaussian processes. He has taught postgraduate courses on statistics, machine learning and deep generative models, and has led data science projects on conservation, health, astronomy, gender studies, and retail.