Speaker
Rose Yu
(UCSD)
Description
AI for Science and Science for AI are usually treated as separate directions. In this talk, I argue they are one entangled system. In one direction, physical structure such as symmetry and conservation makes machine learning reliable and data-efficient, from imposing known symmetries to discovering unknown ones and emulating simulation at scale. In the other, physics can explain the mechanisms of AI itself: the conserved quantities of a neural network's parameter symmetries, in the spirit of Noether's theorem, govern and accelerate training. I will also show how both directions braid together to become the scientific agents that reason like scientists, and discuss the future frontiers of AI and Science.