Conveners
Uncertainty Quantification in ML: Uncertainty Quantification in ML
- Claudius Krause (HEPHY Vienna (ÖAW))
- Alessandra Cappati (Universite Catholique de Louvain (UCL) (BE))
- Riccardo Finotello (CEA Paris-Saclay)
Description
Discussions on UQ techniques and statistical methodology in ML and AI
Welcome speech by the COMETA WG2 coordinators.
The work focuses on generating penalizing configurations for a 2D random field while adhering to manufacturing constraints. Based on an initial well manufactured sample of correlated random field, the proposed methodology involves Principal Component Analysis and subset simulation to incrementally move towards the target density to generate penalizing configurations.
Numerical simulation consists in representing a real experiment using a computer code. Computer models are now essential for simulating and designing complex systems in industrial facilities. Computer simulation is now considered as a third branch for studying phenomena, after theory and real experiments. Its main advantage is to replace costly or infeasible real experiments by numerical...
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...
