PGR Seminar – Carla Davesa Sureda and Gen Li
You are warmly invited to the next PGR seminar:-
Wednesday 29/04/2026
15:00-16:00
JC 1.33A
Speaker 1: Carla Davesa Sureda
Title: Compiling Expressive Planning with Data Types
Abstract: Classical planning relies on PDDL, a language whose expressivity is limited: modellers must encode structured data, numeric reasoning, and counting through propositional workarounds that are often verbose and error prone. In this talk, I present an extension of the Unified Planning framework with high-level modelling constructs: bounded integers, range variables, arrays, sets, and count expressions, together with a set of composable compilers that transform these constructs back into planner-compatible PDDL. This separation allows modellers to work with richer abstractions while preserving access to existing planners. I will describe the compilation pipeline, discuss trade-offs between different compilation strategies, and present results on a range of benchmark domains, such as the 15-Puzzle, Sokoban, and Rush Hour.
Bio: Carla Davesa Sureda is a third-year PhD student in the School of Computer Science, supervised by Prof Ian Miguel and Dr Joan Espasa. Her research focuses on automated planning, in particular on extending the expressivity of planning formalisms through compilation techniques, with connections to constraint programming.
Speaker 2: Gen Li
Title: Visualization of clinical pathways based on sepsis comorbidities
Abstract: Sepsis is a life-threatening condition, yet patients exhibit substantial differences in underlying diseases, disease progression, and clinical outcomes—a phenomenon commonly referred to as clinical heterogeneity. This heterogeneity makes it challenging to identify patient subgroups with similar characteristics and to deliver targeted treatments.
Previous studies have primarily relied on acute physiological measurements and laboratory parameters to stratify patients, often with a focus on outcome prediction. In contrast, comorbidities, as a potential source of heterogeneity, have been less frequently used as primary clustering features. Moreover, few studies have systematically characterized comorbidity co-occurrence patterns within identified subgroups, limiting our understanding of how interactions among comorbidities contribute to sepsis heterogeneity.
Rather than focusing on outcome prediction, this study aims to retrospectively examine how different combinations of diseases influence patients’ clinical pathways and outcomes. To achieve this, we analyse electronic health records of sepsis patients in a large medical database and apply unsupervised learning methods to identify clinically distinct patient subgroups based on comorbidity profiles, and constructed their comorbidity co-occurrence patterns.
Building on this, we further develop a coordinated multi-view visualization dashboard that interactively links phenotypic disease networks with Sankey-based visualization of patient transfer. This system enables cross-perspective analysis from disease co-occurrence structures to clinical pathway patterns, allowing clinicians to explore differences across comorbidity-based subgroups in terms of disease patterns and corresponding care trajectories, thereby supporting the targeted clinical assessment of patients with different disease burdens. Bio: Gen Li is a PhD student in Computer Science, supervised by Dr Areti Manataki (School of Computer Science) and Dr Martin McKechnie (School of Medicine). His research lies at the intersection of health data science and data visualization, currently focusing on disease phenotyping of sepsis patients and the design of interactive clinical pathway visualization systems.