Causal modeling in high-order scenarios: unfolding mechanisms by moving across scales

• Ano de defesa: 2023
• Autor(a) principal: Martinelli, Tiago
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Causal modeling; Emergent phenomena; Fenômenos emergentes; Modelagem causal; Multivariate information theory; Teoria de informação multivariada

The big data era advanced the possibility of studying emergent phenomena in the real world, often occurring by systems with high-order, non-trivial interactions. One of the main questions for these complex systems is to understand how their collective organization influences the dynamic processes. Although such a study is fundamental to developing the policies of controlling dynamical processes from changes in the network structure, in practice, the only information available is multivariate data recorded from variables with unknown topology. Such a scenario can be explored using information theory and causality tools to quantify an individuals influence and infer a causal structure among them. In other words, we can make reverse engineering to obtain a causal model via data. However, a methodology to deal with emergent causes when extracting information is an open question. If not performed correctly, it can compromise basic assumptions in causal modeling resulting in a spurious view of the organization of complex systems. This thesis is dedicated to investigating fundamental problems regarding the capture of emergence phenomena from high-order complex systems joining techniques from causal manipulative approaches and multivariate information theory. Based on our results, we defend a paradigm shift when dealing with multivariate data in causal modeling by considering the task of a system description by moving scales as a fundamental issue instead of a mathematical artifice.