Machine Learning Seminar
Speaker: Phiala Shanahan, College of William & Mary
Title: Machine learning for numerical simulations in nuclear and particle physics
Refreshments available at 2 pm.
Date: Mon, 02 Apr 2018, 2:15 pm – 3:15 pm
Location: 1400 BPS Bldg.
(Dr. Shanahan will also give an informal LQCD seminar at 10 am.)
Many areas of nuclear and particle physics utilize large-scale, hugely computationally demanding, numerical studies of the strong interaction via lattice quantum chromodynamics (LQCD) calculations. Recent advances indicate that multi-scale approaches can provide significant efficiency improvements, and access to new regions of parameter space, if parameters of the LQCD action can be determined from generated datasets. Neural networks provide an efficient solution to this regression task, achieving precise and accurate results even in cases where a principal component analysis fails. This success requires the complex symmetries of LQCD data to be respected within the network structure, a result of the low number of samples and high information content per sample that are features of LQCD datasets. Symmetry-aware machine learning approaches that incorporate the complex invariances of physics problems offer many promising avenues for further study.
Phiala Shanahan completed her Ph.D. at the University of Adelaide, Australia, before moving to MIT as a postdoctoral fellow. She received the American Physics Society Group of Hadronic Physics Thesis Award. In 2017, she was selected as one of Forbes 30 Under 30 Scientists. She is currently an assistant professor at William and Mary and a senior staff scientist at Thomas Jefferson National Accelerator Facility. Her interests span theoretical and experimental nuclear and particle physics.