Computational Science Initiative Event

Machine Learning (ML) capabilities are in a phase of tremendous growth, and there is great opportunity to point these tools toward physical modeling. The challenge is to incorporate domain expertise from traditional scientific discovery into next-generation ML models. We propose to develop new Physics Informed Machine Learning (PIML) algorithms that extend cutting-edge computational and algorithmic ML tools and merge them with physical knowledge in the form of constraints, symmetries, and domain expertise regarding effective degrees of freedom. This PIML methodology is illustrated on the following four enabling examples:


1. Topology and Parameter Estimation in Power Grids [based on arXiv:1710.10727]
2. Creating Turbulent Flows with Deep Learning [based on an APS/DFD2017 abstract]
3. Learning Graphical Models [Science 2018 in print; arXiv:1612.05024]
4. Renormalization of Tensor Networks (Graphical Models) [based on arXiv:1801.01649 and ICML2018 submission]