Federated Graph Neural Networks with Bipartite Embedding for Multi-Objective Facility Location
Zeqi Zheng, Ziqi Wang, Xueming Yan, Yaochu Jin, Shiqing Liu, Qiqi Liu · 2023
Multi-objective facility location problems (MOFLPs) are widely applied in various real-life applications, and an increasing number of deep-learning approaches have been proposed to solve MO-FLPs. However, existing methods often require a substantial amount of data, typically distributed at different data owners who are usually unwilling to share their data. To address this issue, this paper proposes federated graph neural networks with bipartite graph embedding to solve MOFLPs. We formulate the MO-FLP as a bipartite graph and utilize dual graph neural networks to acquire implicit embeddings for nodes and edges, which are trained on their respective data on different clients. Additionally, federated learning with heterogeneous aggregation enables data owners to collaborate on training tasks without sharing raw data, effectively protecting sensitive information related to facility construction and transportation costs. After integrating the models from diverse clients through the central server, we can directly provide the probability distribution of the Pareto set for other instances of the same problem. Experimental results demonstrate that our proposed method can reduce many function evaluations and achieve comparable solution quality compared to state-of-the-art methods, especially in dealing with data of assorted scales.