Be Bayesian, My Friend: My Personal Journey Through Bayesian Learning

ABSTRACT

Scientific problems often involve complex dependencies, multiple sources of uncertainty, and data structures that challenge conventional statistical approaches. Bayesian statistics provides a flexible framework for addressing these challenges by incorporating prior knowledge, modeling complex relationships, and quantifying uncertainty. This seminar will explore how Bayesian hierarchical models, particularly latent Gaussian models, can be used to tackle a wide range of scientific questions. Starting with applications in plant epidemiology and species distribution modeling, we will discuss how spatial and spatio-temporal structures can be incorporated into statistical models, and how computational tools such as Integrated Nested Laplace Approximations (INLA) facilitate Bayesian inference in complex settings.

Special attention will be given to compositional data, where the relative nature of observations introduces additional methodological challenges. We will discuss Bayesian approaches for modeling these data, including methodological developments and computational tools, with applications in ecology, environmental science, and human health.

Finally, we will highlight current challenges and opportunities in Bayesian statistical learning, particularly those arising from high-dimensional and structured data, and illustrate how scientific questions continue to motivate new methodological developments. Throughout the seminar, these topics will be connected through the experiences and lessons learned along a personal research journey, illustrating the interplay between scientific applications, statistical methodology, and computational innovation.

BIOGRAPHY

Joaquín Martínez-Minaya is an Associate Professor in the Department of Applied Statistics and Operational Research and Quality at the Technical University of Valencia (Spain), now a fulbright Visiting Scholar at Harvard University. His research focuses on Applied Bayesian Statistics, with an emphasis on developing statistical methods for the analysis of multivariate, compositional, spatial, and spatio-temporal data. His work combines methodological advances in Bayesian inference with applications in environmental health, ecology, and biomedicine. His research interests include hierarchical models, latent Gaussian models, compositional data analysis, measurement agreement, model assessment, validation, predictive performance, and species distribution modelling. From a computational perspective, he specializes in Bayesian inference using Integrated Nested Laplace Approximation (INLA) and Hamiltonian Monte Carlo methods.