发表机构
University of Nottingham; The University of Hong Kong; Carnegie Mellon University; The Hong Kong Polytechnic University（诺丁汉大学; 香港大学; 卡内基梅隆大学; 香港理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对图结构数据时空预测面临的去中心化异构数据问题，提出结合截断图傅里叶特征等的F²STNet框架，引入FFA调整FedAvg先验，在多个数据集上取得良好预测精度并优化联邦学习的客户端相关指标。
AI 中文摘要
图结构数据的时空预测是交通预测与环境监测的核心，但去中心化且异构的数据既给序列建模也给协同训练带来了复杂问题。我们提出F²STNet，一种联邦预测框架，它结合了截断图傅里叶特征、轻量级对角状态空间时间编码器、图卷积以及公平感知联邦聚合（FFA）。频谱分支揭示图频率结构，状态空间层以序列长度的线性复杂度建模长时序依赖。FFA利用客户端验证损失与递增的公平性调度调整FedAvg先验。在PeMS04、HZMetro和KnowAir上的实验表明，与评估基准相比，该模型预测精度良好；在PeMS04上的联邦实验还提升了最差客户端及客户端分散度指标。
英文摘要
Spatiotemporal prediction on graph-structured data is central to traffic forecasting and environmental monitoring, yet decentralized and heterogeneous data complicate both sequence modeling and collaborative training. We propose F$^2$STNet, a federated forecasting framework that combines truncated graph-Fourier features, a lightweight diagonal state-space temporal encoder, graph convolution, and Fairness-aware Federated Aggregation (FFA). The spectral branch exposes graph-frequency structure, while the state-space layer models long temporal dependencies with linear complexity in the sequence length. FFA adjusts the FedAvg prior using client validation losses and an increasing fairness schedule. Experiments on PeMS04, HZMetro, and KnowAir show favorable forecasting accuracy relative to the evaluated baselines; federated experiments on PeMS04 additionally improve worst-client and client-dispersion metrics.
Comments13 pages
