20/05/2026

Saló de Graus (12:00h)

Mario Alcaide Catalán

Universitat de València

The Gradual Minimum Coverling Location Problem

The Gradual Minimum Covering Location Problem (GMCLP) consists in locating a set of undesirable facilities within a given area. These facilities cover demand of a certain product but have a negative impact over the nearby population. Therefore, the goal is open enough facilities to comply the restrictions while minimizing the population affected by the influence of these undesirable facilities. The first constraint is that enough facilities must be located to cover the demand of the product generated by them. The second constraint forces to ensure that two located facilities are separated by at least a given distance dmin > 0, since being closer may be dangerous. This version of the problem considers different size levels of facilities, each one having different demand coverage and different influence radius. The Gradual Minimum Covering Location Problem was first introduced by Mumtaz Karatas & Levent Eriskin in May of 2021, where they defined the non linear problem and suggested a linear approximation to solve it and one exact formulation. We propose a heuristic algorithm for the original non-linear problem defined by Karatas & Eriskin, based on GRASP and VND. Therefore, we will compare our approach with M. Karatas (2021) proposal. In the numerical experiments, the results show slightly better solutions with an improvement of time to obtain high quality solutions.

28/05/2026

Saló de Graus (12:00h)

Havard Rue

King Abdullah University of Sciences and Technology, Arabia Saudí

Models for graphical/structured and unstructured correlations matrices

The need for models for graphical/structured and unstructured correlations matrices, frequenty comes up in statistical modeling, often in terms of questions like ``I need a prior of a 6 X 6 correlation matrix''. If you additionally have some baseline correlation matrix, there is not that many models to chose from, hardly anyone at all. Before now. In this talk, I willl present our take on this problem, where we construct Penalized Complexity priors for correlation matrices around a baseline model, in the graphical case and in the dense case. (The overall idea is the same, but technicalities are very different in the graphical and dense case.) These models are now available in the R-INLA add-on package graphpcor. This is work with A. Freni-Sterrantino (Turing), E. Krainski (KAUST), J. van Niekerk (Univ Pretoria) and D. Rustand (Univ Bordeaux).

02/06/2026

Seminari del Departament d'Estadística i Investigació Operativa (12:00h)

Inmaculada Arostegui

Euskal Herrriko Unibertsitatea EHU

Modeling recurrence based on previous disease progression with multistate models

This research proposes a two-step statistical approach to better understand disease recurrence, such as reinfection or relapse. In the first step we use multistate models (MSM) to capture the full, complex trajectory of a primary infection, which accounts for all significant health factors and transitions an individual experiences before a potential recurrence occurs.

The second part of the work focuses on the specific risk of recurrence after a fixed point in time, known as a landmark time. It treats death as a competing risk, considering that some patients are no longer at risk of reinfection. Several approaches are proposed and compared, all assuming a Cox model for cause-specific hazards and introducing all baseline and time-dependent covariates at the landmark time.

The proposal is applied to a massive dataset of 400,000 COVID-19 cases in the Basque Country. While the study focused on reinfection, the methodology is designed to be a versatile tool for other acute or chronic illnesses.

This is a joint work with G. Gómez-Mellis and K. Langohr from the Universitat Politécnica de Catalunya.