BayesAI workshop 2026

About

Following the workshops held at Dagstuhl in 2024 and MBZUAI in 2025, the third workshop on Rethinking the Role of Bayesianism in the Age of Modern AI will take place in Edinburgh in October 2026. This workshop will concurrently serve as the inaugural meeting of the International Society for Bayesian Analysis section on Bayesian AI. The gathering will bring together researchers exploring the frontiers of Bayesian machine learning and deep learning in a collaborative atmosphere.

As the scale of this workshop grows, we aim to maintain the diversity of opinions and approaches that have characterised the previous events. Postgraduate students and early career researchers are especially encouraged to participate in this unique opportunity to engage with leaders in the field and discover future research directions.

Purpose

Despite the recent success of AI systems using large-scale deep learning, these systems still fall short in terms of their reliability and trustworthiness. They often lack the ability to estimate their own uncertainty in a calibrated way, encode meaningful prior knowledge, avoid catastrophic failures, and reason about their environments to avoid such failures.

Bayesian deep learning has harboured the promise of achieving these desiderata by combining the statistical foundations of Bayesian inference with the practically successful engineering solutions of deep learning methods. However, compared to its promise, BDL methods often do not live up to expectations in terms of real-world impact.

This workshop aims to rethink and redefine the promises and challenges of Bayesian approaches, elucidate which Bayesian methods might prevail against their non-Bayesian competitors, and identify key application areas where Bayes can shine. The event is planned to promote discussion in a collaborative atmosphere and encourage deep exchange, new ideas, and informal collaboration across intersecting areas of research.

Organising committee

Esmeralda S. Whitammer
University of Edinburgh Sara Wade
University of Edinburgh Vincent Fortuin
TU Nuremberg
Helmholtz AI Konstantina Palla
Spotify Research Theodore Papamarkou
PolyShape
NTU Athens

Registration

Registration is now closed as we have reached the maximum capacity.

Data you submit in the registration form and travel/poster form will be used for communicating with you about the workshop. Personal information will remain in secure cloud storage and will not be shared with anyone outwith the organising committee without your consent, except for purposes directly related to the organisation of the event (e.g., printing badges, providing information to caterers).

By registering, you agree to adhere to the ISBA Code of Conduct.

Refunds wil be issued until 15 September. To cancel your registration, please contact the organisers.

Fees

The generosity of our sponsors allows us to reduce registration fees. Surplus income will be used to provide financial support for participants and to fund the organisation of future Bayesian AI events.

Registration fees are reduced for students and for members of the ISBA Bayesian AI section. It is possible to register as a section member by purchasing or renewing a membership in ISBA.

	Student	Non-student
Section member	£100	£250
Non-section member	£150	£300

Some financial support is available for those for whom fees are a barrier to participation. Please contact the organisers if this applies to you.

Programme

The scientific programme will consist of invited talks, guided discussions, lightning talks, and a poster session for junior participants.

Topics

The workshop will cover a broad range of topics at the interaction of Bayesian methods and modern AI, including:

AI-scale probabilistic reasoning (advances in inference that allow scaling to foundation models);
mathematical underpinnings of generative models (generative modelling in Bayesian AI, deep learning theory in applications to Bayesian methods);
structure-constrained and -informed AI (incorporation of symbolic constraints, structure, and physical priors);
probabilistic and uncertainty-aware methods (model selection and evaluation, uncertainty quantification, applications in trustworthy and green AI);
dynamical systems in probabilistic ML (dynamic process representations, limiting behaviour of Bayesian deep learning models, dynamics-based / diffusion generative modelling);
robustness and adaptation (Bayesian methods for domain adaptation, continual learning, robustness to distribution shift and adversarial attacks);
applications and practical considerations.

Schedule

The below schedule is subject to change and will be filled in over the coming weeks.

	Monday 26.10	Tuesday 27.10	Wednesday 28.10	Thursday 29.10	Friday 30.10
9:00	Registration
9:15	Welcome	Sightseeing activity	TBC	Chris Holmes	Yingzhen Li What would Bayes do for next-gen neural network architecture design?
9:30	Amos Storkey		TBC
9:45			TBC
10:00			TBC		Chris Nemeth Conditioning Gaussian Processes on Almost Anything
10:15	Coffee break		Coffee break and photo	Coffee break
10:30	Coffee break		Coffee break and photo	Coffee break
10:45	Mariia Vladimirova		ISBA section meeting	Thang Bui	Coffee break
11:00			ISBA section meeting		Coffee break
11:15			Tamara Broderick		Teaching Bayesian methods in the age of AI (Ma, Mira, Oates, Rockova, van der Wilk, Wade)
11:30	Michael Gutmann Bayesian design without Bayes			Veronika Rockova
11:45
12:00
12:15	Lunch	Lunch (on your own)	Lunch	Lunch
12:30					Lunch
12:45
13:00
13:15
13:30
13:45	Dmitry Vetrov Hidden Thermodynamics in Deep Neural Networks	Tom Griffiths Bayes and language models, three ways	Mark van der Wilk	Jes Frellsen From Gradients to ‘Reasoning’ Traces: Uncertainty in Language Models
14:00
14:15
14:30	Lightning talks	Isabel Valera	Siddharth Swaroop Adapting knowledge using a Bayesian lens	Julyan Arbel Empowering Bayes with sources of uncertainty
14:45
15:00
15:15		Coffee break	Coffee break	Coffee break
15:30	Coffee break	Coffee break	Coffee break	Coffee break
15:45	Coffee break	Theory and practice: Is Bayes useful in deep learning and applications? (Borovitskiy, Palacios, Rainforth, Rügamer, TBC)	Certification, monitoring, benchmarking, and testing (Harikumar, Panov, Rivasplata, Williams, Fortuin)	Probabilistic machine learning and LLMs / foundation models (Lahlou, Lapata, Siddharth, Storkey, Whitammer)
16:00	Kevin Murphy Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
16:15
16:30
16:45
17:00	Poster session and reception
17:15
17:30
17:45
Late			Dinner at Playfair Library (18:00–)	Social activity (18:30–)

Confirmed participants

Local and travel information

Events will be held at the Informatics Forum (10 Crichton Street, Edinburgh EH8 9AB). The Forum is located on the University of Edinburgh's central campus, adjacent to the historic Old Town and a 15-minute walk from Waverley railway station.

Daily lunch and twice-daily coffee/snacks will be provided. A light social programme is planned, including an opening reception, conference dinner, and local sightseeing activities (details TBA).

Travel, accommodation, and caring support

We plan to offer limited travel and accommodation support, with priority given to junior participants and those from underrepresented groups. Applications for support are now closed.

Support for caring costs is also available thanks to a grant from the Heilbronn Institute. Please contact the organisers if this applies to you.

Contact

For questions to the organisers, please contact Esmeralda Whitammer <esmeralda.whitammer [at] ed.ac.uk>.

Rethinking the Role of
Bayesianism
in the Age of Modern AI

26-30 October 2026

Edinburgh, Scotland, UK