Machine Learning in Physics
Machine Learning for Spectroscopies
Neural network-based algorithms, such as convolutional neural networks, provide us unique opportunities to greatly enhance the amount and accuracy of information we could learn from high dimensional spectroscopies data. X-ray spectroscopies, such as resonant inelastic x-ray scattering (RIXS), and x-ray photo-correlation spectroscopes (XPCS), are powerful tools to investigate the electronic structure, elementary excitations, and non-equilibrium dynamics of quantum materials that could not be reached before. Using neural network-based machine learning algorithms for spectroscopes provides us with unprecedented tools to learn the underlying physics from high-dimensional data.
Manuscript:
​
[1] Hongwei Chen, Sathya R Chitturi, Rajan Plumley, Lingjia Shen, Nathan C Drucker, Nicolas Burdet, Cheng Peng, Sougata Mardanya, Daniel Ratner, Aashwin Mishra, Chun Hong Yoon, Sanghoon Song, Matthieu Chollet, Gilberto Fabbris, Mike Dunne, Silke Nelson, Mingda Li, Aaron Lindenberg, Chunjing Jia, Youssef Nashed, Arun Bansil, Sugata Chowdhury, Adrian E Feiguin, Joshua J Turner, Jana B Thayer, "Testing the data framework for an AI algorithm in preparation for high data rate X-ray facilities", arXiv:2210.10137.
​
[2] Sathya Chitturi, Zhurun Ji, Alexander Petsch, Cheng Peng, Zhantao Chen, Rajan Plumley, Mike Dunne, Sougata Mardanya, Sugata Chowdhury, Hongwei Chen, Arun Bansil, Adrian Feiguin, Alexander Kolesnikov, Dharmalingam Prabhakaran, Stephen Hayden, Daniel Ratner, Chunjing Jia, Youssef Nashed, Joshua Turner, "Capturing dynamical correlations using implicit neural representations", arXiv:2304.03949.
​
​
Crystal Graph Neural Network for Materials Design
Crystal graph convolutional neural network framework (CGCNN) is a powerful method for learning physics properties using crystal structure as the sole input. Trained with hundreds or thousands of structures, CGCNN has been demonstrated to be an effective method for a variety of materials under ambient pressure. We are interested in studying whether CGCNN framework could also be applied to study materials under strain and pressure. Preliminary results for perovskite solar cells under high pressure shows that the trained CGCNN are quite effective in predicting key materials properties for perovskite solar cell materials both at high pressure and with strain.
Manusript
​
[1] Mingkyung Han, Cheng Peng, Feng Ke, Chunjing Jia, Yu Lin,
"Machine Learning Prediction of Perovskite Solar Cell Properties Under High Pressure ", in preparation.
​