07/15/23: Symmetries, Flat Minima, and the Conserved Quantities of Gradient Flow with Bo Zhao

Bo Zhao is a 2nd year PhD student in computer science at UCSD, advised by Rose Yu. Her research focuses on deep learning theory and optimization, with a recent emphasis on the parameter space and dynamics of learning. Today Bo joined us to talk about her recent paper, "Symmetries, Flat Minima, and the Conserved Quantities of Gradient Flow", which was joint work at ICLR with Iordan Ganev, as well as co-authors Robin Walters, Rose Yu, and Nima Dehmamy. This is a really interesting paper which takes an algebraic approach to a problem typically only studied analytically. Bo gave a phenomenal presentation and then we had a really nice discussion with a variety of technical questions. We enjoyed this one a lot and we hope you do too!