Report on ISIPTA 2025
Posted on September 16, 2025 by Justyna Dąbrowska (edited by Alexander erreygers)[ go back to blog ]
The 14th International Symposium on Imprecise Probabilities: Theories and Applications took place on 15-18 July in the most average German city of Bielefeld. The main organisers Frank Riedel, Max Nendel and Jasper De Bock took it upon themselves to greatly expand the topics and interests of this year’s symposium, and made it possible for the attendees – from students, through PhD candidates, to renowned professors from all around the world – to join together in various talks about imprecision and uncertainty.
All the presentations – keynotes, contributed talks, and poster sessions – took place at the ZiF in Bielefeld (that’s short for Zentrum für interdisziplinäre Forschung, in case you’re curious). Each day started with a one-hour keynote, each offering a unique take on uncertainty – from foundational theory to practical applications – highlighting just how many different ways there are to approach the topic of imprecision. The contributed sessions followed the same structure every time: four 9-minute presentations of articles, four 4-minute presentations of abstracts and no intermediate questions, followed by an hour-long intense and detailed discussion by the author’s poster on the topic. On all days, the lunch was served on the bottom floor of ZiF.
It is a (very) difficult task to describe, even somewhat accurately, sessions containing eight presentations each on different topics – all vaguely connected by uncertainty, but ranging from classical probability, through statistics and machine learning, all the way to chemistry and cryptography. Nonetheless, this post will give a briefing of every single presentation – which was a great challenge, and hopefully, a good illustration on how diverse this group have grown over the years, not only nationality-wise (as shown on the map below).
Calculation was based on Book of Contributions, not actual attendees, sorry!
Tuesday
Jan Obłój opened the conference with a talk about Wasserstein Distributionally Robust Optimization. Considering various optimization problems for probability measures and models – from standard regression to deep learning – we learned how to compute Wasserstein distance and describe the models’ uncertainty based on them.
The first session’s talks addressed updating beliefs under dilation and reflection, use of optimal transport for credal sets, possibilistic inferential models and abstract Cauchy problems, while abstract presentations gave us a glimpse of payoffs under model’s ambiguity and stochastic choice.
After lunch, we came for the second round of presentations. We explored imprecise Markov chains, robust Bayesian methods for handling study bias, elicitation techniques for sets of probabilities, as well as distortion of lower probabilities. Abstract contributions focused on quantum cryptography, correlation uncertainty, foundations of Knightian uncertainty and empirical decision problems.
The final session of the day dove into closure properties of aggregation rules, dynamic pricing, credal discrete classification and cycles in credal networks, while short presentations considered two-sampled tests, Cauchy–Schwarz inequalities and a critical reflection of machine learning.
Wednesday
On Wednesday, the day began with the second keynote talk, titled “\(e\)-values, \(e\)-processes, \(e\)-posteriors”, presented by Peter Grünwald. The talk focused on challenges in hypothesis testing – particularly with composite null hypotheses and reliable tracking of Type I error – which highlighted the limitations of standard \(p\)-values. As an alternative, \(e\)-values were introduced, which since their introduction in 2019 have gained significant publicity and are now featured in hundreds of scientific publications.
The following contributed session explored ambiguity preferences, updating rules, precise and imprecise Bayesianism from chemical perspective and distribution-free possibilistic inference, with short abstract presentations considering even more credal networks, more Bayesian variables and risk sharing among many agents.
Group picture captured during one of the very few sunny moments.
After lunch and taking this beautiful picture, the SIPTA general meeting, lead by Jasper De Bock, took place. We had a chance to meet the new members of the executive committee, elected by SIPTA members in the weeks leading to this event.
After six years as president, Jasper De Bock handed over the role to Enrique Miranda, while continuing his involvement as an advisory member. We also welcomed Sébastien Destercke as the new secretary, Alessandro Antonucci as treasurer, and two “new” at-large members: Thomas Augustin and Gert de Cooman. Alexander Erreygers will continue his role as the executive editor, and so will three at-large members: Michele Caprio, Christoph Jansen and Arthur Van Camp.
The day continued with an excursion to a brewery, Brauerei Strate in Detmold, an hour-long drive outside Bielefeld. We had the opportunity to follow the entire beer-making process from start to finish, included with tasting four distinct house-brewed beers (in 40 minutes!). After the tour, we enjoyed a traditional German dinner – together with, naturally, more beer.
Thursday
The third day began with a keynote talk by Carole Bernard titled “Robust Risk Management.” She approached imprecision from a statistical perspective, focusing on quantifying uncertainty in various aggregation functionals – such as variance and expected utility – and on establishing meaningful lower and upper bounds for these quantities.
The first contributed session of the day included a mixture of theory, like rejection functions, quantum mechanics and function-coherence, with more practical views on inference, normality tests, Gaussian mixtures and updating algorithms.
The second session, taking place after a lunch break, included Monte Carlo method for prior-free inference, a revisit of function-coherence, belief change in quantum settings and imprecise versions of conformal predictions. Abstracts added to this mix with ideas on information algebras, dynamic rationality, copula-based estimation and smarter pseudo-labeling in Bayesian networks.
The final session looked at how to handle uncertainty in a smarter and more reliable ways, from improving credal networks and belief updates to making classifiers and financial models more robust. We also touched on how to represent preferences, deal with missing data and reevaluate AI decision in tricky situations.
After a full day of engaging talks and discussions, we came together for the social dinner at Restaurant Hofbräu am Alten Rathaus, where we raised our glasses to celebrate SIPTA’s new president and to thank the organizers for their warm and memorable welcome to Bielefeld.
Friday
The final day began with a keynote talk by Itzhak Gilboa, who introduced the concept of a reasoner – an agent or expert expressing their beliefs – and the rationality axioms that are imposed on them, particularly in light of how we interpret their uncertainty. The talk also walked through aggregation rules and belief updates, and concluded with a discussion of plausible applications of this approach.
The last contributed session opened with a closer look at how local uncertainty models relate to the global ones, how upper transitions behave in the long run, how bringing in varied evidence influence inferences and how to represent conditional prevision through Möbius transformations. The abstracts covered decision-making and hypothesis testing under imprecision, scaling limits in robust optimization and curious dynamics of uncertain random walks.
The conference was closed with an award presentation. First and foremost the IJAR Young Researcher Award 2025, with winners (pictured below from left to right): Siu-Lun Chau (1st prize), Julian Rodemann (2nd prize) and Yusuf Sale (3rd prize). Warm congratulations to all this year’s winners!
The poster prizes went to myself (1st prize), Kathelijne Coussement (2nd prize) and David Nieto-Barba (3rd prize). Thanks for the judges to appreciate my work, and you can take a look at all the winning posters below!
Conclusion
This conference – first one ever for me – was a great deal in many aspects: from learning how many things I have no idea about, through how much I can learn during this short time, and finally, appreciating how much everyone has to offer in this diverse field. Thanks to the organisers, every part of the conference was planned in great detail, and our itinerary – despite already full of contributed sessions – was filled with great activities exploring German culture.
After the symposium took place, the SIPTA community welcomed many new members, increasing by a whooping 10% in size! Today, the society has roughly 150 members.
On a more personal note, beyond all the knowledge gained, the most valuable part of this experience was the sense of community that formed – growing closer over lunches, coffee breaks, and evenings spent together in German bars. It was incredibly motiving to meet so so many open-minded people, and have such a supporting audience throughout all the presentations!
About the author
Justyna Dąbrowska is a first-year PhD student and teaching assistant in the Uncertainty in AI Group, part of the Data and AI cluster at Eindhoven University of Technology in the Netherlands. She is supervised by Arthur Van Camp and Cassio de Campos, and her research focuses on choice functions.