La résilience des réseaux complexes

Authors: Laurence, Edward
Advisor: Dubé, Louis J.; Desrosiers, Patrick
Abstract: Real complex systems are often driven by external perturbations toward irreversible transitions of their dynamical state. With the rise of the human footprint on ecosystems, these perturbations will likely become more persistent so that characterizing resilience of complex systems has become a major issue of the 21st century. This thesis focuses on complex systems that exhibit networked interactions where the components present dynamical states. Studying the resilience of these networks demands depicting their dynamical portraits which may feature thousands of dimensions. In this thesis, three contrasting methods are introduced for studying the dynamical properties as a function of the network structure. Apart from the methods themselves, the originality of the thesis lies in the wide vision of resilience analysis, opening with model-based approaches and concluding with data-driven tools. We begin by developing an exact solution to binary cascades on networks (forest fire type) and follow with an optimized algorithm. Because its practical range is restricted to small networks, this method highlights the limitations of using model-based and highly dimensional tools. Wethen introduce a dimension reduction method to predict dynamical bifurcations of networked systems. This contribution builds up on theoretical foundations and expands possible applications of existing frameworks. Finally, we examine the task of extracting the structural causesof perturbations using machine learning. The validity of the developed tool is supported by an extended numerical analysis of spreading, population, and neural dynamics. The results indicate that subtle dynamical anomalies may suffice to infer the causes of perturbations. It also shows the leading role that machine learning may have to play in the future of resilience of real complex systems.
Document Type: Thèse de doctorat
Issue Date: 2020
Open Access Date: 26 October 2020
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
Description SizeFormat 
36386.pdf3.72 MBAdobe PDFThumbnail
All documents in CorpusUL are protected by Copyright Act of Canada.