Events 2026/27

Events 2026/27

Tuesday 10 November 2026, 13:00-14:00, IF room G.03

Speaker: Chris Williams

Title: Use Bayesian Paired Tests with a ROPE to Improve the Comparison of Machine Learning Models; and How Should We Review Papers in the Age of AI?

Abstract: Model comparison in machine learning can be much improved by using \emph{paired testing}, i.e.\ comparing  the predictions of methods A and B on each (common) test example. Due to the limitations of null hypothesis significance testing, a Bayesian approach is recommended, including the use of the region of practical equivalence (ROPE).

 

Tuesday 29 September 2026, 11:00, IF room G.07

Joint IML / RAS-CDT talk

Speaker: Masashi Sugiyama (RIKEN/The University of Tokyo)

Title: Recent Advances in Reward Modeling for Reinforcement Learning

Abstract: Reinforcement learning (RL) has achieved remarkable success in robotics, games, and language model post-training, where reward signals are essential for effective agent training. In this talk, I will present our recent research on advanced reward modeling. This includes robustifying RL through transfer and weakly supervised learning, coping with interval-based rewards, and exploring diverse reward aggregation frameworks that move beyond the traditional discounted sum.

Biography: Masashi Sugiyama received his Ph.D. in Computer Science from the Tokyo Institute of Technology, Japan, in 2001. After serving as an assistant and associate professor at the same institute, he became a professor at the University of Tokyo in 2014. Since 2016, he has also served as the director of the RIKEN Center for Advanced Intelligence Project. His research interests include theories and algorithms of machine learning, such as weakly supervised learning, distribution shift adaptation, and reinforcement learning. He was a keynote speaker at ICLR 2023, and is currently an ICML board member and a NeurIPS advisory board member.

Friday 28 August 2026, 10:00, IF room G.03

Speaker: Futoshi Futami (University of Osaka and RIKEN AIP, https://sites.google.com/view/futoshifutami/home)

Title: A Short Tutorial on PAC-Bayes: Basic Ideas and Recent Topics

Abstract: This talk provides a short tutorial on the basic ideas of PAC-Bayes theory. I will explain its main principles and its connections to Bayesian inference and variational Bayes. I will then discuss several recent topics, including information-theoretic generalisation bounds, applications to variational autoencoders, and extensions to sequential decision-making problems.

Biography: Futoshi Futami is an associate professor at the graduate school of engineering science, the University of Osaka , and is also affiliated with the RIKEN AIP. His research focuses on the theoretical foundations of machine learning, including PAC-Bayes theory, information-theoretic analysis, and methods for uncertainty quantification.

Friday 28 August 2026, 11:00, IF room G.03

Speaker: Wenkai Xu (University of Warwick, https://warwick.ac.uk/fac/sci/statistics/staff/academic-research/xuw/)

Title: A Kernel Nonconformity Score for Multivariate Conformal Prediction

Abstract: Conformal prediction has recently attracted attention in post-learning inference. In this talk, we ll start by revisiting some basics on the topic. We then introduce a recent development to construct conformal region for multi-variate outputs. A Multivariate Kernel Score (MKS) will be discussed and we show that the proposed score resembles the Gaussian process posterior variance, unifying Bayesian uncertainty quantification with the coverage guarantees of frequentist-type. Moreover, we demonstrate that the MKS reduces the volume of prediction regions significantly compared to ellipsoidal baselines and conclude by some ongoing and future directions in the topic.

Biography: Wenkai Xu is currently an assistant professor of statistics and machine learning at the department of statistics, University of Warwick. He is also a visiting researcher at Graduate School of Frontier Sciences, University of Tokyo; and Tuebingen AI Center. His research interests includes hypothesis testing, Stein's method, network analysis, and information-theoretical approaches for statistics and machine learning.

This article was published on