On 12 December 2025, we had our annual in-person event that was jointly organised with the RSS Applied Probability Section. The event was also available online and it should be noted that more individuals were registered to attend the event online rather than in-person.
The event featured three talks and two coffee breaks that enabled discussions between the participants and the presenters.
Below are short descriptions of the talks presented at the event.
Fix-A-Shortcut: Gradient-Based Shortcut Mitigation for Causal Representation Learning in Healthcare
Sonali Parbhoo, Imperial College London
Dr Parbhoo is an Assistant Professor and leader of the AI for Actionable Impact (AI4AI) lab at Imperial College London.
Dr Parbhoo presented work on how to handle spurious correlations, namely shortcuts, that are irrelevant to the problem in-hand. Dr Parbhoo presented a method developed by her group based on gradient descents, named Fix-A-Shortcut (FAS). The proposed approach has been shown to reduce shortcuts significantly compared to standard approaches used in the field, like regularisation approaches. The application of the proposed methodology was illustrated on several healthcare applications.
The work presented by Dr Parbhoo is available at:
- Hong, H. et al, 2025 “Do Regularization Methods for Shortcut Mitigation work as Intended?”, Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, PMLR 258:3349-3357
Designing experiments on networks
Vasiliki Koutra, King’s College London
Dr Koutra, a Lecturer in Statistics at King’s College London, introduced the audience to the task of trial design in the presence of interference that controls the variation of the subjects when the main interest is on treatment estimation. More specifically on how to design experiments for network data. Dr Koutra presented both theoretical contributions as well as a number of emerging applications with networked data. For example, Dr Koutra presented an analysis of a subset of the Facebook Network for estimating the effect of two treatments and how the different types of approaches perform. An agricultural experiment was also presented where the aim was to quantify the impact of neighbouring plots in field trials. Some of the work presented by Dr Koutra is available through the following research articles:
- Koutra, V., Gilmour, S.G., and Parker, B.M. (2021). ”Optimal Block Designs for Experiments on Networks,” Journal of the Royal Statistical Society, Series C, 70, 596–618.
- Koutra, V., Gilmour, S.G., Parker, B.M., and Mead, A. (2023).”Design of Agricultural Field Experiments Accounting for Complex Blocking Structures and Network Effects,” Journal of Agricultural, Biological, and Environmental Statistics, 28, 526–548.
Bayesian inference for tail risk extrapolation in time series
Simone Padoan, Bocconi University
Dr Padoan is an Associate Professor at the Department of Decision Sciences at Bocconi University. Dr Padoan presented a Bayesian model for estimating the risk in the case of extreme events that have not yet be seen. Dr Padoan firstly introduced the audience to the main concepts of extreme events, and continued by presenting a novel Bayesian inference approach of marginal-tail quantile-based risks. Dr Padoan presented the likelihood, proposed prior and posterior distributions of the proposed Bayesian approach with the theoretical work that supported the relevant choices. Alongside numeric experiments that showcased the performance of the method, Dr Padoan showed examples of emerging applications where they applied the proposed approach, including the analysis of real interest rates. Dr Padoan’s presented work can be accessed at:
- Carl, D.L, Padoan, S.A., and Rizzelli, S. (2025).”Accurate Bayesian inference for tail risk extrapolation in time series”, arXiv:2510.14637