Jul 2026· ISPRS International Journal of Geo-Information· 0 citations· 50 references
TL;DR
Meta-FedGeo is introduced, a federated learning framework that integrates meta-learning and spatiotemporal transformers to address key challenges in urban GeoAI for smart cities and advances GeoAI toward scalable, adaptive, and practical applications in smart city environments.
Abstract
This study introduces Meta-FedGeo, a federated learning framework that integrates meta-learning and spatiotemporal transformers to address key challenges in urban GeoAI for smart cities. Data streams in smart city environments are inherently non-stationary and heterogeneous, limiting the adaptability of traditional federated learning approaches. Meta-FedGeo overcomes these limitations through a hybrid centralised–decentralised architecture that pre-trains a global model using meta-learning to capture cross-city spatiotemporal patterns and dynamically refines it through federated updates. The framework incorporates performance-aware client selection and temporally weighted aggregation to enhance model robustness and convergence. To model complex urban dynamics, the proposed system employs a Spatio-Temporal Transformer (ST-Transformer). In addition, an Uncertainty-Calibrated Decision Engine (UCDE) is introduced to align model predictions with accessibility and urban planning constraints. Unlike static federated methods, Meta-FedGeo can dynamically identify and filter malicious or low-quality clients using local validation loss, while Shapley value-based mechanisms support efficient and fair knowledge transfer across distributed nodes. To clarify the scope of the present study, Meta-FedGeo is reported as a partially implemented research prototype: the ST-Transformer backbone, the meta-learning initialisation, the validation-loss-based client filtering and the temporal-weighted aggregation were implemented and evaluated on partitioned real-world datasets, whereas the Shapley-value contribution assessment, the Lightweight Data Harmonisers (LDHs) and the UCDE are presented as architectural components with proof-of-concept implementations whose full empirical validation is identified as future work. The framework is designed for seamless integration with existing urban infrastructure without requiring major modifications. Experimental results using real-world urban datasets partitioned into non-IID federated clients indicate improved predictive performance and faster convergence relative to the federated baselines considered here. Overall, Meta-FedGeo advances GeoAI toward scalable, adaptive, and practical applications in smart city environments.
Air quality, as a global issue, impacts people’s health and daily life. It requires precise prediction and monitoring for sustainable urban management. Conventional centralized air quality prediction methods are limited by data privacy, high communication costs, and low scalability in distributed environments. Federated learning offers a solution that collaboratively trains global models without sharing raw data. The passage conducted a comprehensive literature analysis to categorize FL applications in air quality research into four technical streams: Probabilistic Graphical Models (Federated Bayesian Networks) for causal inference; Time-series Deep Learning (integrating LSTM/CNN with FedAvg/FedProx) for temporal pattern extraction; Spatio-temporal Graph Neural Networks (GC-LSTM) for capturing complex spatial dependencies; and Multi-model Ensemble with Transfer Learning for heterogeneous client adaptation. Nevertheless, there are significant gaps that remain in terms of model interpretability and the ability to generalize across climate variations. All in all, this article provides a relatively comprehensive overview of the application of federated learning in air quality forecasting and monitoring. It will assist researchers in advancing this field in the future.
Yuhao Wu· Mathematical Modeling and Al...· 0 citations
Recently, the integration of Internet of Things (IoT) into buildings has sparked a technological revolution known as the Internet of Building Things (IoBT). Such new paradigm connects devices, sensors, and systems for intelligent control. Particularly, IoBT has advanced energy forecasting and enabled real-time data analysis and predictive optimization for smarter and more efficient energy management. This paper proposes FETRA, a FEderated TRansformer with Attention-based client weighting and adaptive FedProx regularization for smart building energy prediction. FETRA addresses data heterogeneity through dynamic client importance assignment, captures long-term temporal dependencies via specialized attention mechanisms, and enhances convergence stability under non-IID distributions. Experimental evaluation on the ASHRAE dataset with 100 heterogeneous buildings demonstrates that the Informer model achieves superior performance with 94.18% prediction accuracy, MSE of 892.45, RMSE of 29.87, and $\mathrm{R}^{2}$ of 0.94, outperforming baseline techniques. In addition, the edge-fog-cloud architecture adapted in FETRA enables privacy-preserving distributed energy forecasting while maintaining computational efficiency. Results confirm FETRA's effectiveness in balancing prediction accuracy, system robustness, and scalability for real-world deployment in decentralized building energy management systems.
Jessica Al Achy, Hassan Harb, Abdallah Makhoul· Annual International Compute...· 0 citations
This work proposes FedTT, a federated spatio-temporal learning framework for cross-city traffic knowledge transfer that consistently outperforms 18 state-of-the-art baselines, achieving improved prediction accuracy while maintaining strong empirical resistance to attacks.
Zhihao Zeng, Ziquan Fang, Yuting Huang et al.· Proceedings of the 32nd ACM...· 0 citations
Traffic congestion significantly impacts safety and urban livability in smart cities, motivating the development of accurate Traffic Flow Prediction (TFP) systems. Traditional deep learning approaches typically rely on centralized training, which is difficult to scale in distributed Internet of Things (IoT) environments. To address these limitations, decentralized paradigms such as Local Learning (LL) and Federated Learning (FL) enable on-device training and collaborative model updates while preserving data locality. For real-time TFP, the inherently non-stationary nature of traffic data necessitates continuous model adaptation, making online federated learning essential for scalable and collaborative deployment. However, this setting remains challenging because traffic data are typically non-IID across clients, with local patterns varying significantly across locations and devices. This paper presents an exploratory study of online federated learning for TFP on resource-constrained IoT devices. Using a GRU-based network as a common backbone, the Online Federated Learning paradigm is benchmarked relative to Centralized and Local Learning as reference baselines. A performance evaluation was conducted by evaluating RMSE and MAE on the PEMS-BAY dataset. Robustness to non-IID data is further assessed using FedProx and SCAFFOLD. Results show that LL achieves the lowest prediction error, whereas FL degrades as the number of local epochs increases, and non-IID mitigation strategies provide limited improvements under low-latency constraints. Overall, online federated learning is a viable approach for real-time TFP, but its performance is highly sensitive to client heterogeneity.
M. Pizzolante, A. Shumba, T. Montanaro et al.· Annual International Compute...· 0 citations
Experiments on controlled synthetic data and the Q-Traffic real-world dataset demonstrate that the proposed framework improves predictive accuracy, cross-client stability, and robustness under heterogeneous federated traffic scenarios.
Zhi-Cheng Wang, Tao Zhang, Yi-Meng Zhu et al.· Applied Sciences· 0 citations
Prototype-based knowledge sharing effectively mitigates data and model heterogeneity in federated learning (FL) by exchanging class-level semantic information. However, existing methods typically assume all local prototypes are equally reliable. Consequently, low-quality prototypes from heterogeneous models or dynamic clients can contaminate the global aggregation, leading to a vicious cycle of noise accumulation and performance degradation. To address this, we propose FedLEAF, a Federated Learning framework with server-side proactive Evaluation and clientside Adaptive Fusion. Specifically, the server employs an Adaptive Learning Prototypes (ALP) network to dynamically evaluate prototype reliability and generate learnable aggregation weights, ensuring that highquality prototypes exert a primary influence on the global model. Meanwhile, the client utilizes a Historical Consistency Fusion (HCF) strategy to selectively absorb global knowledge by assessing its consistency with locally maintained historical prototypes. Extensive experiments on standard datasets demonstrate that FedLEAF achieves effective improvements in model accuracy and robustness compared to existing methods.
Zhiyuan Zhu, Si-Yi Deng, Dapeng Wu et al.· 2026 International Conferenc...· 0 citations