AutoFed: Personalized Federated Traffic Prediction via Adaptive Prompt
Summary: arXiv:2512.24625v2 Announce Type: replace-cross
Introduction
Accurate traffic prediction is a cornerstone for the development of Intelligent Transportation Systems (ITS), which encompass ride-hailing services, urban road planning, and vehicle fleet management. Despite its significance, traffic prediction faces substantial challenges, particularly concerning data privacy. Most existing prediction methodologies depend on local training, which often leads to data silos and a lack of collaborative knowledge sharing.
The Challenge of Federated Learning
Federated Learning (FL) has emerged as a potential solution, enabling privacy-preserving collaborative training. However, standard FL encounters difficulties with the non-independent and identically distributed (non-IID) problem that exists among different clients. This limitation has given rise to Personalized Federated Learning (PFL), a paradigm that aims to address these issues. Despite its promise, existing PFL frameworks often require adaptation for traffic prediction tasks, such as specialized graph feature engineering, data processing, and tailored network architecture design.
Limitations of Current Approaches
A significant constraint in many previous studies is their dependency on hyper-parameter optimization across datasets—information that is frequently unavailable in real-world contexts. This reliance hampers the practical deployment of these predictive models, leaving a gap in effective traffic prediction solutions.
Introducing AutoFed
To tackle these challenges, we propose AutoFed, a novel PFL framework designed specifically for traffic prediction. AutoFed eliminates the necessity for manual hyper-parameter tuning, thus simplifying the implementation process. Drawing inspiration from prompt learning, AutoFed introduces a federated representor that utilizes a client-aligned adapter to distill local data into a compact, globally shared prompt matrix.
How AutoFed Works
This shared prompt matrix conditions a personalized predictor, allowing each client to harness cross-client knowledge while preserving local specificity. The framework thus strikes a balance between global cooperation and individual client needs, enhancing the efficacy of traffic prediction models.
Performance Evaluation
Extensive experiments conducted on real-world datasets demonstrate that AutoFed consistently outperforms existing methods across a variety of scenarios. The results highlight the framework’s ability to adapt to different traffic conditions and its robustness in handling diverse data distributions.
Conclusion
In summary, AutoFed presents a significant advancement in the field of traffic prediction through its innovative use of PFL and prompt learning techniques. By addressing the key limitations of prior approaches, AutoFed offers a promising pathway for the development of accurate, privacy-preserving traffic predictive models.
Further Information
The code for AutoFed is available at GitHub – AutoFed.