F2STNet: Fair and Federated Spectral-Temporal Modeling for Graph Forecasting
2026-08-10 • Machine Learning
Machine Learning
AI summaryⓘ
The authors developed F²STNet, a method to predict things like traffic or pollution over time and across connected locations using decentralized data. Their approach combines techniques that capture patterns in graph structures and long sequences efficiently, plus a way to fairly combine information from different data sources in a federated setting. Tests on real-world datasets show their method predicts well and treats all data contributors more fairly. This helps in scenarios where data is spread out and owned by different parties who collaborate without sharing raw data.
spatiotemporal predictiongraph-Fourier transformstate-space modelsgraph convolutionfederated learningfairness in machine learningsequence modelingtraffic forecastingenvironmental monitoring
Authors
Jiayi Zhang, Jinfeng Xu, Hewei Wang, Siyuan Cen, Haidong Huang, Yiyao Zhan, Zheyu Chen, Jinjiang You, Ai Jian, Edith C. H. Ngai
Abstract
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.