.png){kind=icon}
.png){kind=icon}
Return
Source localization in complex networks with optimal observers based on maximum entropy sampling
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2024.124946.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Despite significant progress in source localization, challenges persist in identifying optimal observers for nonlinear propagation models and locating sources during the early stages of information spread with one observation snapshot. Additionally, leveraging unobserved nodes to enhance localization accuracy remains an open question. In this paper, we introduce maximum entropy sampling theory to optimize observers, and innovatively consider the unobserved nodes and further propose a heuristic method for inferring the state of unobserved nodes. By combining maximum entropy sampling with inferred unobserved node states, we present an improved label back-propagation method for source localization, establishing a comprehensive framework for source localization in complex networks. Extensive simulations demonstrate the ability of our framework to locate sources in the early stages. Moreover, our findings reveal that observers selected through maximum entropy sampling concentrate on fringe nodes. Remarkably, neglecting unobserved node states results in similar performance of source localization across observation methods, whereas the combination of maximum entropy sampling and inferred unobserved node states significantly enhances the performance of source localization, say, the AUROC is increased by about 0.1. Furthermore, compared with heterogeneous network structures, our method has better performance of source localization on homogeneous networks.
Keywords:
Complex networks
Source localization
Information spread
Maximum entropy sampling
Optimization observers
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
View more organizations