Balázs Csanád Csáji Delivered a Keynote Lecture at the 25th EYSM Conference of the Bernoulli Society

Balázs Csanád Csáji, a senior researcher at the Research Laboratory of Engineering & Management Intelligence (EMI) of HUN-REN SZTAKI, was invited as a keynote speaker to the 25th European Young Statisticians Meeting (EYSM), organized by the Bernoulli Society for Mathematical Statistics and Probability, held in Vilnius, Lithuania, on July 7-10, 2026. Alongside delivering his keynote speech on robust uncertainty quantification in machine learning, he also participated as an expert panelist in a discussion panel addressing the challenges of artificial intelligence for science and society.

During the meeting, hosted by the Faculty of Mathematics and Informatics of the Vilnius University, Balázs Csanád Csáji delivered his keynote talk titled "Robust Uncertainty Quantification: from Resampling and Ranking to Stochastic Bandits." His presentation addressed how to reliably quantify uncertainty while minimizing rigid, unrealistic structural and distributional assumptions. The talk introduced distribution-free, non-asymptotic confidence region constructions based on resampling and ranking techniques, such as the Sign-Perturbed Sums (SPS) method. As a key highlight, he presented the Resampled Median-of-Means (RMM) estimator, which constructs confidence intervals for symmetric (even heavy-tailed) distributions with optimal shrinkage rates without requiring prior knowledge of moments. When applied to stochastic bandit problems, this approach yields a data-driven algorithm (RMM-UCB) that guarantees near-optimal regret and outperforms most state-of-the-art bandit algorithms on hard stochastic bandit problems (i.e., when the suboptimality gap is small and the reward distributions are heavy-tailed).

In addition to his keynote, Balázs Csanád Csáji was invited to join a panel of international experts to discuss the "Challenges of AI for Science and Society." The panel explored the rapid integration of AI into scientific workflows and its impact on education and research, with a particular emphasis on the question of the correctness and verifiability of mathematical proofs generated by AI systems.