A2TTA: Anchored-and-Agile Test-Time Adaptation for Evolving Traffic Sensor Networks
A new framework called A2TTA (Anchored-and-Agile Test-Time Adaptation) addresses the challenge of adapting traffic forecasting models to evolving sensor networks. Traditional models assume fixed sensor graphs, but real-world networks continuously change due to road construction and shifting mobility patterns. A2TTA handles two key issues: topology expansion from new sensors and connections, and temporal shifts varying in time scale and stability. The framework efficiently adapts pretrained models during deployment without retraining from scratch. This research was published on arXiv with ID 2607.25875.
Key facts
- A2TTA stands for Anchored-and-Agile Test-Time Adaptation.
- It targets evolving traffic sensor networks.
- Topology expansion introduces new sensors and connections.
- Temporal shifts include long-term and short-term changes.
- The framework adapts pretrained models during deployment.
- Traditional models assume fixed sensor graphs.
- Real-world networks evolve due to road construction and mobility patterns.
- The paper is on arXiv with ID 2607.25875.
Entities
Institutions
- arXiv