Preface

Great progresses have been made by researchers for the mutual devleopment between physically-based models (mechanism models) and Artificial intelligence (AI for short). Several typical milestones represent different genres, while reaching out the same purpose. To clarify each of them, from personal perspectives, groups are roughly made for better understanding of tags and categories in the following blogs.

Scientific machine learning

Three separate aspects of scientific benchmarking that apply in the context of ML benchmarks for science, namely, scientific ML benchmarking, application benchmarking and system benchmarking¹.

• Scientific ML benchmarking. This is concerned with algorithmic improvements that help reach the scientific targets specified for a given dataset.
  • • SciNet²: discvering simple physical concepts between distances and angles in the earth-moon system.
  • • Physics-informed neural networks³: solve and discover low-dimension nonlinear partial differential equations.
  • • Deep operator networks: a nested structure in learning operators (relationships between the input and output function).
• Application benchmarking. This aspect of ML benchmarks is concerned with exploring the performance of the complete ML application.
  • hybrid models⁴: combineing concepts from machine learning and physically-based models, by parameterizing or replacing part of mechanisms with machine learning, in order to entitle ML with interpretability.
    • • hybrid hydrological model⁵
  • causal deep learning⁶:
• System benchmarking. This is concerned with investigating performance effects of the system hardware architecture on improving the scientific outcomes/targets.

Reference

1. Scientific machine learning benchmarks. (2022). https://doi.org/10.1038/s42254-022-00441-7. ↩

2. Discovering physical concepts with neural networks. (2020). https://doi.org/10.48550/arXiv.1807.10300. ↩

3. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. (2019). https://doi.org/10.1016/j.jcp.2018.10.045. ↩

4. Deep Learning and Process Understanding for Data-Driven Earth System Science. (2019). https://doi.org/10.1038/s41586-019-0912-1. ↩

5. Hybrid modeling: fusion of a deep learning approach and a physics-based model for global hydrological modeling. (2020). https://doi.org/10.5194/isprs-archives-XLIII-B2-2020-1537-2020. ↩

6. Causal deep learning. (2023). https://arxiv.org/abs/2303.02186. ↩