Uncertainty Quantification in Machine Learning

Monday 27 Jan 2025, 10:00 → 12:05 Europe/Paris

Alessandra Cappati (Universite Catholique de Louvain (UCL) (BE)), Claudius Krause (HEPHY Vienna (ÖAW)), Riccardo Finotello (CEA Paris-Saclay)

In recent years, the landscape of scientific research has been dramatically reshaped by advancements in AI and machine learning (ML). These tools have enabled us to process vast amounts of data and uncover complex patterns with unprecedented efficiency. However, as we delve deeper into these realms, the importance of understanding and quantifying uncertainty in our computations has become increasingly apparent.

Uncertainty Quantification (UQ) is not just about acknowledging the limitations of our models and data, but also about harnessing this understanding to make our predictions more robust and our conclusions more reliable. It is a vital component of scientific rigour, enabling us to navigate the complexities of real-world systems with confidence.

In this meeting, we aim to foster a cross-disciplinary dialogue on the challenges and opportunities presented by UQ. We will discuss how it can enhance our understanding in particle physics, improve the reliability of mathematical models, and inform the development of AI and ML tools.

10:00 → 12:05 Uncertainty Quantification in ML

Discussions on UQ techniques and statistical methodology in ML and AI

Conveners: Alessandra Cappati (Universite Catholique de Louvain (UCL) (BE)), Dr Claudius Krause (HEPHY Vienna (ÖAW)), Dr Riccardo Finotello (CEA Paris-Saclay)

10:00 Introduction

Welcome speech by the COMETA WG2 coordinators.

Speakers: Alessandra Cappati (Universite Catholique de Louvain (UCL) (BE)), Dr Claudius Krause (HEPHY Vienna (ÖAW)), Dr Riccardo Finotello (CEA Paris-Saclay)

10:05 Uncertainty Quantification methodology for science and industry

Speaker: Dr Clément Gauchy (CEA Paris-Saclay)

10:35 Q&A Session

10:45 Optimal Experimental Designs of Fracture Toughness Test Campaigns Under Uncertainties

Speaker: Anthony Quintin (CEA Paris-Saclay)

11:15 Q&A Session

11:25 Using amplitudes for accuracy and uncertainty estimation in ML

Neural networks in LHC physics have to be accurate, reliable, and controlled. We first show how activation functions can be systematically tested with KANs. For reliability and control, we learn an uncertainty together with the target amplitude over phase space. While systematic uncertainties can be described by a heteroscedastic loss, a comprehensive learned uncertainty requires Bayesian networks or repulsive ensembles. We compute pull distributions to show that the learned uncertainties are calibrated correctly.

Speaker: Nina Elmer (Heidelberg University)

11:55 Q&A Session