Joint Seminar: Generative Cascades of Multivariate Extremes

Abstract: In this talk, Professor Miguel de Carvalho, GAIL Fellow will first motivate the notion of cascades of extremes—as recently introduced by de Carvalho, Ferrer & Vallejos (2026, Extremes)—such as a tsunami followed by an earthquake or a cascade of stock market losses; a neural model will be used to discuss and illustrate the main concepts and ideas. He will then argue that the sequential nature of cascading extremes makes them particularly appealing for the setting of generative extremes, and introduce an generative framework grounded on extreme value theory that synthesizes cascades through causal attention. Within this framework, a causal transformer maps the history of extremes into a latent state that governs direction, magnitude, and inter-event timing. Simulated and real data examples will be used to illustrate the scope of the proposed approaches. 

Short bio: Miguel de Carvalho is Professor and Chair of Statistical Data Science at the University of Edinburgh. He is elected fellow of GAIL, co-director of the ECFI, and a member of the Council of the International Statistical Institute. Miguel’s research interests include, inter alia, extreme value theory, Bayesian analysis, and the interfaces between statistics and AI. He has been an AE for a variety of top tier journals such as Bayesian Analysis, The American Statistician, The Annals of Applied Statistics, and the Journal of the American Statistical Association. Miguel co-chaired the international conference EVA 2021 in Edinburgh, co-edited the special issue Bridging Heavy Tails & AI, and is the leading editor of the Handbook of Statistics of Extremes.