Keynote
Artificial intelligence from a statistical perspective
By Benjamin Guedj
Research Director at Inria, France and Professor at the University College London, UK
As artificial intelligence continues to transform science, society, and industry, the need for principled and interpretable learning algorithms has never been greater.
In this keynote, I will offer a statistical perspective to artificial intelligence, highlighting the crucial role that statistical thinking plays in understanding and designing intelligent systems. Drawing on contributions from statistical machine learning, I will explore key ideas in generalisation theory—how and why models trained on data perform well on unseen inputs—and present recent advances in PAC-Bayesian theory that provide tight, information-rich generalisation guarantees.
I will also discuss modern approaches to uncertainty quantification, from hypothesis testing to conformal prediction that yield distribution-free, finite-sample guarantees.
I will reflect on the role of Bayesian statistics in modelling uncertainty and supporting robust, data-efficient learning, paving the way for sustainable (or frugal) artificial intelligence.
Biography
Prof Dr Benjamin Guedj is a Research Director at Inria in France and a Professor of Machine Learning and Foundational Artificial Intelligence, University College London in UK.
He also serves as a Turing Fellow at The Alan Turing Institute in UK. Benjamin conducts research in theoretical machine learning. He holds a PhD in mathematics from Sorbonne Université in France and focuses on statistical learning theory, PAC-Bayes, computational statistics, mathematics of deep learning, among other topics.
Benjamin is the founder and scientific director of the Inria London programme, a Franco-British joint research lab between Inria and UCL.
Benjamin is a co-director of the ELLIS UCL Unit, a Young Leader and Trustee of the Franco-British Council, and a Knight of the Order of the Academic Palms of the French Republic.