Haixu Wu

Logo

wuhaixu98@gmail.com

About Me

I am currently a Postdoc in the Computational Design and Fabrication Group at MIT CSAIL, supervised by Prof. Wojciech Matusik. Previously, I received my Ph.D. and Bachelor’s degree from School of Software, Tsinghua University, under the supervision of Prof. Mingsheng Long.

My research centers on deep learning and scientific machine learning, aiming to develop scaling principles for physical systems–principles that remain largely absent from current foundation models and are essential for unlocking the full potential of AI for the physical world. With a particular focus on physical and dynamical systems, my work spans three main directions:

Google Scholar / GitHub / CV

Education

Experience

Highlights

Apr 2026 We have released GeoPT, which offers a promising way to scale up physics simulation and has been awarded as Best Paper at ICLR 2026 Workshop on Foundation Model for Science.
Feb 2026 Neural solver backbone Transolver has been extended to a third version, Transolver-3, which supports PDE solving on geometries with over 100-million cells.
Oct 2025 We have released FlashBias, an extremely optimized kernel for fast computation of attention with bias, which enables 1.5x speedup for Pairformer in AlphaFold 3. Try FlashBias!
Feb 2025 We have released Neural-Solver-Library as a simple and neat codebase for developing neural PDE solvers. Welcome to try this library and join the research in solving PDEs.
Jun 2023 Unified forecasting model for worldwide stations (Corrformer) was published as the Cover Article in Nat. Mach. Intell. and was awarded as Youth Outstanding Paper Nomination of WAIC 2024.
May 2023 Time series model (TimesNet) was selected as most influential papers by Paper Digest.
Apr 2022 Spatiotemporal backbone (PredRNN) was awarded as ESI Highly Cited Paper and Hot Paper.
Dec 2021 Time series backbone (Autoformer) was selected as most influential papers by Paper Digest.

Selected Publications [Full List]

Journal Articles

  1. Interpretable Weather Forecasting for Worldwide Stations with a Unified Deep Model
    Haixu Wu, Hang Zhou, Mingsheng Long#, Jianmin Wang#
    Nature Machine Intelligence 2023 / PDF / Code / Slides (Cover Article, WAIC Youth Outstanding Paper Nomination)

  2. PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning
    Yunbo Wang*, Haixu Wu*, Jianjin Zhang, Zhifeng Gao, Jianmin Wang, Philip S. Yu, Mingsheng Long#
    TPAMI 2022 / PDF / Code (ESI Highly Cited Paper, ESI Hot Paper)

Conference Proceedings

  1. GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training
    Haixu Wu*, Minghao Guo*, Zongyi Li, Zhiyang Dou, Mingsheng Long, Kaiming He, Wojciech Matusik
    ICML 2026 / Code / Project Page (Best Paper at ICLR 2026 Foundation Model for Science Workshop)

  2. FlashBias: Fast Computation of Attention with Bias
    Haixu Wu, Minghao Guo, Yuezhou Ma, Yuanxu Sun, Jianmin Wang, Wojciech Matusik, Mingsheng Long#
    NeurIPS 2025 / PDF / Code / Slides / Poster

  3. Unisolver: PDE-Conditional Transformers Are Universal PDE Solvers
    Hang Zhou*, Yuezhou Ma*, Haixu Wu#, Haowen Wang, Mingsheng Long#
    ICML 2025 / PDF / Code / Slides / Competition (3rd in NeurIPS FAIR Universe Competition)

  4. Transolver++: An Accurate Neural Solver for PDEs on Million-Scale Geometries
    Huakun Luo*, Haixu Wu*, Hang Zhou, Lanxiang Xing, Yichen Di, Jianmin Wang, Mingsheng Long#
    ICML 2025 / PDF / Code / Slides

  5. RoPINN: Region Optimized Physics-Informed Neural Networks
    Haixu Wu, Huakun Luo, Yuezhou Ma, Jianmin Wang, Mingsheng Long#
    NeurIPS 2024 / PDF / Code / Slides

  6. Transolver: A Fast Transformer Solver for PDEs on General Geometries
    Haixu Wu, Huakun Luo, Haowen Wang, Jianmin Wang, Mingsheng Long#
    ICML 2024 / PDF / Code / Slides / Poster (Spotlight Paper)

  7. SimMTM: A Simple Pre-Training Framework for Masked Time-Series Modeling
    Jiaxiang Dong*, Haixu Wu*, Haoran Zhang, Li Zhang, Jianmin Wang, Mingsheng Long#
    NeurIPS 2023 / PDF / Code / Slides (Spotlight Paper)

  8. Solving High-Dimensional PDEs with Latent Spectral Models
    Haixu Wu, Tengge Hu, Huakun Luo, Jianmin Wang, Mingsheng Long#
    ICML 2023 / PDF / Code / Slides

  9. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis
    Haixu Wu*, Tengge Hu*, Yong Liu*, Hang Zhou, Jianmin Wang, Mingsheng Long#
    ICLR 2023 / PDF / Code / Slides (PaperDigest Most Influential Paper)

  10. Flowformer: Linearizing Transformers with Conservation Flows
    Haixu Wu, Jialong Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long#
    ICML 2022 / PDF / Code / Slides

  11. Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy
    Jiehui Xu*, Haixu Wu*, Jianmin Wang, Mingsheng Long#
    ICLR 2022 / PDF / Code / Slides (Spotlight Paper)

  12. Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
    Haixu Wu, Jiehui Xu, Jianmin Wang, Mingsheng Long#
    NeurIPS 2021 / PDF / Code / Slides (PaperDigest Most Influential Paper)

  13. MotionRNN: A Flexible Model for Video Prediction with Spacetime-Varying Motions
    Haixu Wu*, Zhiyu Yao*, Jianmin Wang, Mingsheng Long#
    CVPR 2021 / PDF / Appendix / Code / Slides

* Equal Contribution, # Corresponding Author

System and Applications

Experience

Selected Awards