CAREER: Understanding Invariant Convolutional Neural Networks through Many Particle Physics

点击翻译按钮获取中文摘要

Large scale computing in engineering and the physical sciences, in addition to new experimental methods, are generating massive amounts of high dimensional distributed data, and interpreting, analyzing, and using this data is a fundamental problem facing science. Machine learning algorithms extract, interpolate, and extrapolate information from these data sets. However, the complexity of modern data requires new algorithmic paradigms that are capable of parsing intricate and subtle patterns across numerous scales. This CAREER award will make significant strides in developing these new machine learning paradigms, focusing on multiscale systems of interacting bodies that are used to model molecules, materials, and drugs, amongst others.More specifically, this CAREER award will facilitate an integrated scientific and educational program at the interface of mathematics, deep learning, many particle physics and data science. Scientifically, it will develop a mathematical theory of invariant convolutional neural networks based upon (wavelet) scattering transforms, and seek to interpret the mathematical properties of these scattering transforms through the lens of many particle systems. The research will facilitate fast simulations of quantum many particle systems, thus opening new research avenues in quantum chemistry, materials science and drug discovery. Intertwined with these research efforts is a plan to increase the participation of under-represented students in data science. The centerpiece of this part of the project is a data science themed "alternate spring break," meant to introduce undergraduate students to data science research, which will be integrated with existing programs at Michigan State University to maximize impact.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.