Machine Learning with Physics Knowledge for Prediction: A Survey

Joe Watson,Chen Song,Oliver Weeger,Theo Gruner,An T. Le,Kay Hansel,Ahmed Hendawy,Oleg Arenz,Will Trojak,Miles Cranmer,Carlo D'Eramo,Fabian Bülow,Tanmay Goyal,Jan Peters,Martin W. Hoffman

2024-08-19

Abstract:This survey examines the broad suite of methods and models for combining machine learning with physics knowledge for prediction and forecast, with a focus on partial differential equations. These methods have attracted significant interest due to their potential impact on advancing scientific research and industrial practices by improving predictive models with small- or large-scale datasets and expressive predictive models with useful inductive biases. The survey has two parts. The first considers incorporating physics knowledge on an architectural level through objective functions, structured predictive models, and data augmentation. The second considers data as physics knowledge, which motivates looking at multi-task, meta, and contextual learning as an alternative approach to incorporating physics knowledge in a data-driven fashion. Finally, we also provide an industrial perspective on the application of these methods and a survey of the open-source ecosystem for physics-informed machine learning.

Machine Learning,Numerical Analysis,Computational Physics

What problem does this paper attempt to address?

The paper aims to address the issue of how to combine physical knowledge with machine learning methods to improve predictive models. Specifically, this review paper focuses on the following aspects:

1. **Methods of Integrating Physical Knowledge**: Integrating physical knowledge at the architectural level through objective functions, structured predictive models, and data augmentation.
2. **Data-Driven Approaches**: Exploring data-driven methods such as multi-task learning, meta-learning, and contextual learning as another way to integrate physical knowledge.
3. **Industrial Application Perspective**: Providing prospects for the application of these methods in industry and investigating the open-source ecosystem.

The core contributions of the paper are:
- Identifying and communicating the challenges in learning physical information, providing pathways for future research directions.
- Offering key guidance by placing the concept of learning physical information within the broader context of machine learning and connecting knowledge from different engineering fields.
- Establishing a refined classification system of methods, providing a structured framework for different approaches.
- Evaluating the relevance and application scenarios of these methods in industry.
- Predicting future trends to bridge the gap between theoretical research and practical applications, enabling physical information machine learning to more effectively solve real-world problems.