.png){kind=icon}
.png){kind=icon}
Return
Multi-Source Localization Based on Graph Representation Learning and Bayesian Optimization
.png){kind=icon}
PRE
AI
DOI：10.1109/TNSE.2025.3644931.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Source localization, the inverse problem of diffusion processes, is crucial for tracking social rumors, identifying epidemic spreaders, and detecting computer viruses. Multi-source localization based on snapshot observation has garnered significant attention due to its low cost and ease of acquisition. However, challenges such as ill-posedness and heavy dependence on diffusion models hinder effective solutions. Existing methods often rely on deterministic techniques that require searching the entire graph space, struggle to effectively encode topological information, and are limited to a single diffusion model. To address these limitations, we propose Source Localization based on Representation Learning and Bayesian Optimization (SL-RLBO), a generic framework that quantifies source uncertainty via Monte Carlo simulation. Specifically, we first develop a novel algorithm to simultaneously estimate diffusion parameters and time from a single snapshot. Then, we utilize a multi-source reverse infection algorithm to identify candidate sources and leverage graph representation learning techniques to capture latent topological features. Finally, we formulate an objective function applicable to various diffusion models and efficiently optimize it using Bayesian optimization. Extensive experiments and case studies conducted on two synthetic and six real-world datasets show that SL-RLBO consistently outperforms four state-of-the-art baselines across different diffusion models, reducing error distance by an average of 18.94%.
Keywords:
Bayesian optimization
graph representation learning
multi-source localization
parameter estimation
Journal
I
IF:
7.9
Papers:
2.5K
Citations:
10.0K