# Integrating Machine Learning with Physics-Based Modeling

**Authors:** Weinan E, Jiequn Han, Linfeng Zhang  
**arXiv:** [2006.02619](https://arxiv.org/abs/2006.02619)  
**Date:** 2020-06-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2006.02619v1/

## Key points

- Machine learning has the potential to revolutionize scientific research
- Challenges must be addressed before this potential can be realized
- Integrating machine learning with physics-based modeling is a key issue of interest
- Authors Weinan E, Jiequn Han, and Linfeng Zhang provide guidelines for integrating machine learning with physics-based modeling
- Critical issues in developing machine learning-based physical models include imposing physical constraints and obtaining optimal datasets
- Addressing these challenges can enhance interpretability and reliability of models
- Modern machine learning techniques have been successful due to practical considerations outlined by the authors
- A concurrent machine learning framework is introduced for blending machine learning with physics-based modeling seamlessly
- Examples from molecular dynamics and moment closure of kinetic equations demonstrate how integration can lead to more accurate scientific models
- The article raises questions about the implications and new frontiers that may emerge from successfully integrating machine learning into scientific modeling

## Summary

Machine learning has emerged as a powerful tool with the potential to revolutionize scientific research. However, before this can become a reality, several challenges must be addressed. One key issue of interest is how to effectively integrate machine learning with physics-based modeling to create new, interpretable, and reliable physical models. In their article "Integrating Machine Learning with Physics-Based Modeling," authors Weinan E, Jiequn Han, and Linfeng Zhang delve into this topic. The authors begin by outlining general guidelines for integrating machine learning with physics-based modeling. They then highlight two critical issues in developing machine learning-based physical models: imposing physical constraints and obtaining optimal datasets. By addressing these challenges, researchers can enhance the interpretability and reliability of the models generated through this integration. In addition to discussing practical considerations, the authors provide a straightforward explanation for why modern machine learning techniques have been successful. They also introduce a concurrent machine learning framework necessary for seamlessly blending machine learning with physics-based modeling. To illustrate their points, the authors use examples from molecular dynamics and moment closure of kinetic equations. These examples showcase how integrating machine learning with physics-based modeling can lead to more accurate and insightful scientific models. The article concludes with a forward-looking discussion on the implications of successfully integrating machine learning into scientific modeling. It raises questions about where this integration will take us and what new frontiers will emerge once machine learning becomes an integral part of scientific research. Overall, the article provides valuable insights into the challenges and opportunities associated with combining machine learning and physics-based modeling in pursuit of enhanced scientific understanding.