Welcome to the NSF AI Planning Institute for Data-Driven Discovery. Physics applications of AI have led to some of the most exciting recent breakthroughs, from the synthesized radio imaging of the event horizon of a black hole (which used Machine Learning for image reconstruction) to explorations of the functioning of the human genome (Deep Learning for regulatory genomics and cellular imaging).
Scientists at CMU in departments such as Machine Learning and Statistics actively collaborate with scientists in the Department of Physics because of the opportunity for each field to spur development in the other. This collaboration is now accelerating, with weekly interactive seminars, now including AI scientists, astrophysicists, particle physicists, and an emerging team of biophysicists.
Our past work together spanning two decades, the recent excitement generated by these no-jargon seminars, and our recent NSF award emboldens us to propose a joint Physics/AI Planning Institute. The aim is to bring cutting edge methods from AI into a broad range of physics areas, to rapidly propagate successful methods from one field of Physics to another thereby avoiding replication of effort, and to facilitate back-transfer from the data-rich sub-fields of physics to AI development.
The Planning phase will focus on areas where CMU scientists are already leaders, in which there are existing strong collaborations between physicists and AI researchers, and where rapid advances are being made: astrophysics, subatomic physics, and biophysics. Applying AI will lead to significant advances in the areas of dark energy and galaxy formation; new ways of extracting information about the Higgs bosons and anomalies in gluon physics; and enhanced understanding of biological networks and predictions for cancerous tissues. Benefits in the other direction are clear as well: physics provides complex use cases and profound problems that motivate AI researchers to advance foundational AI.