⭐ ST-TTC is a method for exploring the real-time calibration of models in the face of open environment dynamic spatio-temporal distribution shifts during the Test-Time Computing Phase.
🚩 News (Sep. 2025): ST-TTC's code, training details, and inference details are fully open source! Try to improve on this! 😊
🚩 News (Sep. 2025): ST-TTC has been accpeted by NeurIPS 2025 with Spotlight! ✅
Spatio-temporal forecasting is crucial in many domains, such as transportation, meteorology, and energy. However, real-world scenarios frequently present challenges such as signal anomalies, noise, and distributional shifts. Existing solutions primarily enhance robustness by modifying network architectures or training procedures. Nevertheless, these approaches are computationally intensive and resource-demanding, especially for large-scale applications. In this paper, we explore a novel test-time computing paradigm, namely learning with calibration, ST-TTC, for spatio-temporal forecasting. Through learning with calibration, we aim to capture periodic structural biases arising from non-stationarity during the testing phase and perform real-time bias correction on predictions to improve accuracy. Specifically, we first introduce a spectral-domain calibrator with phase-amplitude modulation to mitigate periodic shift and then propose a flash updating mechanism with a streaming memory queue for efficient test-time computation. ST-TTC effectively bypasses complex training-stage techniques, offering an efficient and generalizable paradigm. Extensive experiments on real-world datasets demonstrate the effectiveness, universality, flexibility and efficiency of our proposed method.
-
Large-Scale Scenario: The experimental code for some settings of RQ2 in this article is in the large_scale_scenario file. Please refer to the README.md in the folder for related experiments.
-
Small-Scale Scenario: The experimental code for some settings of RQ1 and RQ2 in this article is in the small_scale_scenario file. Please refer to the README.md in the folder for related experiments.
-
OOD Learning Setting: The experimental code for the first part of the scenario of RQ3 in this article is in the ood_learning_setting file. Please refer to the README.md in the folder for related experiments.
-
Continual Learning Setting: The experimental code for the second part of the scenario of RQ3 in this article is in the continual_learning_setting file. Please refer to the README.md in the folder for related experiments.
🌟 If you find the ST-TTC helpful in your research, please consider to star this repository and cite this paper:
@inproceedings{chen2025stttc,
title={Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting},
author={Wei Chen and Yuxuan Liang},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
year={2025}
}
We also welcome to cite our previous work:
@inproceedings{chen2025eac,
title={Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting},
author={Wei Chen and Yuxuan Liang},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025}
}
We appreciate the following GitHub repos or Websites a lot for their valuable code, data and efforts.
This project is licensed under the Apache-2.0 License.