Machine-Guided Discovery of a Real-World Rogue Wave (瘋狗浪) Model

台北天文館，AI 找到如何預測巨浪的公式，2023 年 12 月 02 日 

Dion Häfner, Johannes Gemmrich, Markus Jochum, Machine-Guided Discovery of a Real-World Rogue Wave Model,  Proceedings of the National Academy of Sciences (2023), 120(48). (arXiv, data, code)

Machine learning has had a transformative impact on predictive science and engineering. But due to their black-box nature, better machine learning models do not always lead to greater human understanding, the first goal of science. We show how this can be overcome by using machine learning to transform a vast database of wave observations into a human-readable equation for the occurrence probability of rogue waves—rare ocean waves that routinely damage ships and offshore structures. This equation can be analyzed and incorporated into the research canon. Our work demonstrates the potential of causal analysis, machine learning, and symbolic regression to drive scientific discovery in a real-world application.

3 steps in Figure 1:

1. A-priori analysis of causal pathways that leads to a set of presumed causal parameters (Section 1).
2. Training an ensemble of regularized neural network predictors, and parsimony-guided model selection based on causal invariance (Section 2).
3. Distillation of the neural network into a concise mathematical expression via symbolic regression (Section 3).