Balázs Csanád Csáji Opens the Publication Awardees’ Lecture Series at HUN-REN SZTAKI

Last Friday marked the debut of the "Publication Awardees" lecture series at HUN-REN SZTAKI. Of the three Awardees, the first presentation of the series was delivered by Balázs Csanád Csáji on the topic of "Robust Uncertainty Quantification."

Uncertainty is an inherent part of natural and social sciences, engineering, industry, finance and healthcare. This arises from factors such as limited knowledge, measurement errors, modeling biases, computational constraints, and the intrinsic variability of the systems we study. Consequently, Uncertainty Quantification (UQ) is a fundamental requirement for building trustworthy models and ensuring robust decision-making, which is especially important in safety-critical environments.

The lecture focused on stochastic UQ methods that move beyond restrictive distributional assumptions and asymptotic results, which often fail in real-world scenarios. A UQ approach is considered "robust" when it minimizes underlying statistical and structural assumptions.

The talk demonstrated how ideas from nonparametric statistics, such as resampling and ranking, can lead to efficient, actionable UQ methods providing distribution-free and non-asymptotic guarantees. Beyond the high-level conceptual overview, the presentation illustrated these ideas through applications, including constructing distribution-free confidence regions for parametric and nonparametric regression models.