OPTIMAL: Optimisation and Post-Bayesian Inference in Machine Learning

05 May 2026, Tangier, Morocco

Welcome!

The aim of probabilistic machine learning is to find accurate representations of our uncertain beliefs about the world and use them to make better informed decisions. This workshop brings together post-Bayesian approaches to inference and optimisation-based perspectives on uncertainty and decision-making. Post-Bayesian methods address the limitations of classical Bayesian inference by developing alternative inferential principles that remain robust in modern machine-learning settings, where standard modelling assumptions may be violated. Complementing this view, optimisation-based approaches treat inference and decision-making as problems of optimising functionals of probability distributions, providing a unifying framework for both learning probabilistic representations and acting upon them. This workshop welcomes all theoretical and methodological work on how best to represent, find and use probabilistic beliefs about the world.

Call for Papers

We invite short papers submissions on all theoretical and methodological work on how best to represent, find and use probabilistic beliefs about the world.

Please submit your paper through OpenReview

Submission details

Submissions should be formatted using the AISTATS LaTeX style. Papers are limited to 4 pages (excluding references). The review process will be double-blind. Accepted contributions will be presented as posters, and selected works will also be highlighted as contributed talks.

Key dates

The important deadlines are listed below.

Fri, Jan 30, 2026 Submissions open
Fri, Mar 6, 2026 Submissions close
Fri, Mar 27, 2026 Decisions announced
Tue, May 5, 2026 Workshop begins

Keynote speakers

All presenters are listed in alphabetical order.

Chris Oates (Newcastle University)
Detecting Model Misspecification in Bayesian Inverse Problems via Variational Gradient Descent

Clémentine Chazal (CREST / ENSAE, Paris)
A Computable Measure of Suboptimality for Entropy-Regularised Variational Objectives

Geoff Pleiss (University of British Columbia)
Perfect Bayesian Optimization is Hard; “Good Enough” Bayesian Optimization is Easy

Pierre Alquier (ESSEC Asia-Pacific)
Empirical PAC-Bayes Bound for Markov Chains

Schedule

Time Activity
09:00–09:10 Opening Remarks
09:10–09:40 Invited Talk - Pierre Alquier
09:40–10:00 Contributed Talks
Towards E-Value Based Stopping Rules for Bayesian Deep Ensembles
Emanuel Sommer, Rickmer Schulte, Sarah Deubner, Julius Kobialka, David Rügamer
10:00–10:30 Coffee
10:30–11:00 Invited Talk - Clémentine Chazal
11:00–12:00 Contributed Talks
A Distributional Optimisation Perspective on Ensemble Methods
Yan Lin, Congye Wang, Zheyang Shen, Matthew A Fisher, Chris J. Oates

Sequential Updating of Predictively Oriented Posteriors
Zheyang Shen, Gerardo Duran-Martin, Chris J. Oates

Even More Guarantees for Variational Inference in the Presence of Symmetries
Lena Zellinger, Antonio Vergari
12:00–12:30 Invited Talk - Chris Oates
12:30–14:00 Lunch
14:00–14:30 Invited Talk - Geoff Pleiss
14:30–15:30 Panel Discussion
15:30–18:00 Poster Session

Accepted Papers

A Predictive View on Streaming Hidden Markov Models
Gerardo Duran-Martin

Addressing Stochastic Rising Bandits with Thompson Sampling
Marco Fiandri, Francesco Trovò, Alberto Maria Metelli

Axiomatizing Tempered Bayesian Updating via Local Likelihood Transformations
Yutong Zhang, Yaoran Yang

Bayesian inference with sources of uncertainty: a confidence-weighted approach to sparsity
Rafael Mouallem Rosa, Julyan Arbel, Hien Duy Nguyen

Closed-Form Reward Centroids for Inverse Reinforcement Learning
Filippo Lazzati, Alberto Maria Metelli

Encoding Inductive Biases in Simulation-based inference
Ben Riegler, Vincent Fortuin

Guiding Posterior Exploration with Optimizer-Derived Geometry
Moritz Schlager, Emanuel Sommer, Thomas Möllenhoff, David Rügamer

Implied Likelihoods in Linear Amortised Bayesian Methods
Samuel Power

Indirect Query Bayesian Optimization with Integrated Feedback
Mengyan Zhang, Shahine Bouabid, Cheng Soon Ong, Seth Flaxman, Dino Sejdinovic

Learning with Embedded Linear Equality Constraints via Variational Bayesian Inference
Matthew Marsh, Benoit Chachuat, Antonio Del rio chanona

Occam’s Razor is Only as Sharp as Your ELBO
Ethan Harvey, Michael C Hughes

Optimal information deletion and Bayes’ theorem
Hans Montcho, Håvard Rue

Pandora’s Regret: Decision-Aligned Evaluation for Sequential Search
Gerardo Flores, Ashia C. Wilson

Regularization Effects in Variational Training of Transformers
Yi Han, Jonathan Wenger, John Patrick Cunningham

Rethinking Probabilistic Circuit Parameter Learning
Anji Liu, Zilei Shao, Guy Van den Broeck

Rethinking Trust Region Bayesian Optimization in High Dimensions
Wei-Ting Tang, Joel Paulson

Robust and Adaptive Bayesian Contextual Bandits in Heavy-Tailed and Piecewise-Stationary Environments
Gianluca Palmari, Alvaro Cartea, Fayçal Drissi, Gerardo Duran-Martin

Robust Bayesian Experimental Design under Misspecification
Hany Abdulsamad, Sahel Iqbal, Christian A. Naesseth, Takuo Matsubara, Adrien Corenflos

Robust Obedience in Information Design for Bayesian Congestion Games
Yuwei Hu, Bryce Ferguson

Scalable Uncertainty Quantification for Black-Box Density-Based Clustering
Nicola Bariletto, Stephen Walker

Teacher Forcing as Generalized Bayes: Optimization Geometry Mismatch in Switching Surrogates for Chaotic Dynamics
Andre Herz, Daniel Durstewitz, Georgia Koppe

The Benefits of Sampling in Unregularized Variational Training of Deep Neural Networks
Juraj Marusic, Jonathan Wenger, Beau Coker, John Patrick Cunningham

Unlocking the Secrets of Perturbation Methods for End-to-End Prediction and Optimization
Kyle Heuton, Michael C Hughes

When Does Feel-Good Thompson Sampling Fail Under Approximate Posteriors?
Emile Timothy Anand, Sarah Liaw

Organisers