About this tutorial
Data science is playing an ever-increasing role in physics. While some departments have offered courses, many of the examples are in the context of social science and other disciplines. In the second part of this tutorial we will cover several advanced applications of data science to physics. We will cover physics-informed neural networks (PINNs), which incorporate the physical properties of partial differential equations into the structure and training of neural networks. Examples of PINNs will focus on geophysical and climate-related problems, but PINNs can be applied in other contexts, such as Density Functional Theory. We will also cover image segmentation: how do we automatically detect, segment, and classify objects in a wide variety of images? Examples will be presented from biophysical contexts, but the methods can be applied to many other fields of physics. We will discuss Bayesian Optimization/Active learning, which leads to better sampling of measurements, and can be applied to both theory and experiment. Our final topic will be Sparse Identification of Nonlinear Dynamics (SINDy), which deals with learning equations from experimental data. Data Science is playing an ever-increasing role in physics. While some departments have offered courses, many of the examples are in the context of social science and other disciplines. In the second part of this tutorial we will cover several advanced applications of data science to physics. We will cover physics-informed neural networks (PINNs), which incorporate the physical properties of partial differential equations into the structure and training of neural networks. Examples of PINNs will focus on geophysical and climate-related problems, but PINNs can be applied in other contexts such as Density Functional Theory. We will also cover image segmentation: how do we automatically detect, segment, and classify objects in a wide variety of images? Examples will be presented from biophysical contexts, but the methods can be applied to many other fields of physics. We will discuss Bayesian Optimization/Active learning, which leads to better sampling of measurements, and can be applied to both theory and experiment. Our final topic will be Sparse Identification of Nonlinear Dynamics (SINDy), which deals with learning equations from experimental data.
Topics covered in this tutorial include:
- Physics informed neural networks
- Image segmentation
- Bayesian optimization
- SINDy/symbolic regression
Speakers
- Ching-Yao Lai, Stanford University
- Jan Funke, Howard Hughes Medical Institute Janelia Research Campus
- Arpan Biswas, Oak Ridge National Laboratory and University of Texas
- Chris Amey, Brandeis University
Organizers
- William Ratcliff, National Institute of Standards and Technology
- Talat Rahman, University of Central Florida