Quantum-Enhanced Parameter-Efficient Learning for Typhoon Trajectory Forecasting

Published in IEEE INFOCOM 2025 – IEEE International Conference on Computer Communications, 2025

This paper introduces a Quantum-Enhanced Parameter-Efficient Learning framework for typhoon trajectory forecasting, marking the first application of quantum machine learning (QML) in this domain. Key features include:

  • Use of Quantum Parameter Adaptation (QPA) to train a hybrid Attention-based Multi-ConvGRU model with significantly fewer trainable parameters.
  • Quantum-Train (QT) methodology enables parameter generation during training using quantum neural networks, with no quantum hardware required at inference.
  • Demonstrated accuracy and efficiency on large-scale atmospheric datasets, with strong potential for scalable and energy-efficient climate modeling.

The framework showcases how QML can drive breakthroughs in disaster forecasting while improving accessibility and sustainability in high-performance computing.

Figure: Third-place project at Pasqal Climate & Biodiversity Challenge 2024—QPA model for storm trajectory prediction.

Recommended citation: Liu, Chen-Yu; Chen, Kuan-Cheng; Chen, Yi-Chien; Chen, Samuel Yen-Chi; Huang, Wei-Hao; Huang, Wei-Jia; & Chang, Yen-Jui. (2025). "Quantum-Enhanced Parameter-Efficient Learning for Typhoon Trajectory Forecasting." Proceedings of IEEE INFOCOM 2025 – IEEE International Conference on Computer Communications.
Download Paper