When large individual losses are rare, the empirical tail of the loss distribution becomes unstable and often underestimated. This workshop offers an applied and interactive case study focusing on heterogeneity and rare events. Participants will engage hands-on with data to explore modelling choices, assess tail risk, and compare approaches under limited information. Teams will experiment with a range of modelling techniques, such as alternative severity families, inclusion of covariates, regularization methods, extreme-value extensions, and strategies like resampling versus parametric tail modelling. They will also examine fair evaluation criteria in small-sample contexts. After a short industry framing of the main challenges (pricing, risk measurement, and portfolio management), participants will work in teams on a synthetic dataset calibrated to realistic large-loss scenarios.

The rise of artificial intelligence (AI), primarily through large language models (LLMs), has already dramatically reshaped teaching, learning, analysis, and scholarship. In this workshop, we will cover through hands-on tutorials how LLMs can be (and are being) used throughout this process, with a focus on the research workflow. This will include topics ranging from using LLMs for general search, brainstorming/discussion, and editing/feedback on writing to more advanced topics such as "vibe coding" (i.e. programming via natural language) and the creation of fully agentic systems for research and analysis. It will aim to provide practical, actionable advice for implementing these tools within scientific workflows.

This workshop will be run jointly with the Statistical Education Workshop  "AI and LLMs - Education Perspective" in the morning to provide a common, foundational introduction to working with LLMs before splitting off in the afternoon to cover independent material. Both workshops will meet in MDCL-1009 in the morning.

This one-day tutorial will present the mathematical theory of Markov chain convergence, including such concepts as random walks, recurrence and transience, stationary distributions, reversibility, etc.  It will describe Brownian motion and diffusions as continuous limits of discrete Markov chains.  It will then apply this knowledge to Markov chain Monte Carlo (MCMC) algorithms, explaining their convergence and efficiency from a theoretical viewpoint.  It will explain how MCMC algorithms can converge to diffusions, and how to use that fact to optimise their performance. Depending on time, it may also discuss how to optimise tempering algorithms, and/or prove convergence of adaptive MCMC algorithms.  No background is assumed beyond basic undergraduate-level probability theory and mathematical reasoning, and perhaps a bit of familiarity with MCMC.

Generative AI is driving change in classrooms across every subject, and statistics classrooms are no exception. This interactive workshop invites participants to critically examine the potential and pitfalls of integrating AI into their teaching practice. The focus will be on practical use, ethical considerations, and data privacy. Sessions will include discussion and showcase current use-cases to motivate the hands-on activities where participants will use AI tools to create teaching resources and learn to programmatically interact with generative AI models. Both cloud-based and locally hosted large language models (LLMs) will be used. Code templates, instructions, and links to resources will be provided; your exploration of AI does not need to end when the workshop does. In fact, it may have just started!

This workshop will be run jointly with the Data Science and Analytics Section Workshop "AI and LLMs - Research Perspective" in the morning to provide a common, foundational introduction to working with LLMs before splitting off in the afternoon to cover independent material. Both workshops will meet in MDCL-1009 in the morning.

Machine-learning methods are increasingly used in national statistical offices and survey organizations, mainly to produce predictions at different stages of a survey. This course covers how to use those predictions for valid finite-population inference. We discuss model-assisted estimation and imputation for item nonresponse; for unit nonresponse, we examine what ML changes for inverse-probability weighting, what is currently justified, what is not, and why. Topics include standard and doubly robust estimators, variance estimation using cross-fitting, and asymptotically valid confidence intervals. Practical issues such as hyperparameter tuning, and weight trimming will also be discussed. By the end, participants will have an up-to-date toolkit for valid inference when ML predictions are used in surveys, and a clear view of the key open problems for unit nonresponse.

Synthetic datasets are increasingly used in biostatistics and health research to enable data sharing and protect patient privacy. While these datasets often resemble the original data closely, analyzing them requires care : standard inferential procedures may no longer apply, and ignoring the variability introduced during synthesis can lead to misleading conclusions. This half-day workshop offers a practical introduction to analyzing synthetic data with appropriate methodology, such as specific combining rules, and will include hands-on implementation in R. Examples of some existing synthetic datasets in the biostatistics domain will be provided.