|
Abstract
摘要 |
Complex networks have provided powerful tools for modeling different systems in the real world and uncovering the relation between network structures and functional properties, especially regarding system vulnerability. The talk includes two parts, which concentrate on two important methods in complex network models and system vulnerability.
The first part focuses on interdependent networks. In fully interdependent random networks, an initial node attack invokes a cascade of iterative failures that may lead to a total collapse of the whole system in a form of abrupt first order transitions. When the system is near criticality, the giant component decreases slowly in a plateau form and the number of iterations in the cascade diverges. In our work, we interpret this plateau behavior by describing the cascading failures as growing failure trees. This work helps us in designing strategies for preventing and mitigating catastrophic collapses.
The second part focuses on the application of correlation-based network methods on different real systems. In climate networks, we define links based on cross-correlation from global sites with climate time series. We propose a new method to reduce the indirect effects in the observed networks, and we find the optimal propagation paths behind the observed teleconnections. This work provides a new tool to explore the global scale impacts of major climate events (e.g. El Niño). In communication networks, we generate multi-layer correlation-based networks from packet loss/downtime data in several mobile networks. We measure the shared failure events using the mean correlation and degree distributions. In this work, we interpret the observed power-law/lognormal degree distributions by building an event impact model. We also capture the boundary between "sparse enough" and "dense enough" systems through the mean degree peaks. Correlation-based networks provide a general framework to detect dependency relations in many different temporal-spatial systems, which thereby improves our understandings of system vulnerability under event impacts. |
