M$^2$FedAQI: Multimodal Federated Learning for Air Quality Prediction on Heterogeneous Edge Devices

In three linesM²FedAQI introduces a lightweight multimodal federated framework for decentralized Air Quality Index (AQI) prediction across heterogeneous edge devices. The system fuses visual and tabular data through feature modulation-based fusion. Evaluated on PM25Vision and TRAQID datasets, it achieves 11% accuracy improvement, 3.53% AUC gain, 12.2% F1-score increase, and 18% R² improvement over baselines.

Summary generated by Claude — human-verified