From Cities to Series: Complex Networks and Deep Learning for Improved Spatial and Temporal Analytics

• Ano de defesa: 2021
• Autor(a) principal: Souza, Gabriel Spadon de
• 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: Artificial intelligence; Ciência de redes; Inteligência artificial; Network science; Séries temporais; Sistemas urbanos; Time series; Urban systems
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-01092021-104851/

The relationship between different entities is a property that can be represented as a graph, structured sets formed by entities (i.e., vertices) and relationships (i.e., edges). Graphs have often been used to answer questions about the interaction between entities from the real world by analyzing their vertices and edges (i.e., the graphs topology). On the other hand, complex networks are known to be graphs of non-trivial topology, capable of representing human phenomena such as cities urbanization, peoples movement, and migration, besides epidemic processes. However, graph theory and network science, the research fields that oversee the study of graphs and complex networks, have also been traversed in the realm of artificial intelligence, in which the analysis of the interaction between different entities is transposed to the internal learning process of algorithms. In this sense, this thesis introduces complex networks and supervised learning (classification and regression) techniques to improve understanding of human phenomena inherent to street networks, pendular migration, and pandemics progression through computational analysis and modeling. Accordingly, we contribute with: (i) techniques for identifying inconsistencies in the urban plan while tracking the most influential vertices; (ii) a methodology for analyzing and predicting links in the scope of human mobility between cities through machine learning algorithms; and (iii) a new neural network architecture capable of modeling dynamic processes observed in spatial and temporal data with applications on different domains. These results reiterate the potential of graphs and complex networks in solving problems related to analyzing human phenomena and modeling their evolutive processes across space and time when used together with articial intelligence learning algorithms.