BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:James Halverson (Northeastern University) DTSTART:20201119T183000Z DTEND:20201119T193000Z DTSTAMP:20260422T212922Z UID:nhetc-special/1 DESCRIPTION:Title: Neural Networks and Quantum Field Theory\nby James Halverson (No rtheastern University) as part of Special NHETC Seminar\n\nLecture held in Zoom.\n\nAbstract\nWe propose a theoretical understanding of neural netwo rks in terms of Wilsonian effective field theory. The correspondence relie s on the fact that many asymptotic neural networks are drawn from Gaussian processes\, the analog of non-interacting field theories. Moving away fro m the asymptotic limit yields a non-Gaussian process and corresponds to tu rning on particle interactions\, allowing for the computation of correlati on functions of neural network outputs with Feynman diagrams. Minimal non- Gaussian process likelihoods are determined by the most relevant non-Gauss ian terms\, according to the flow in their coefficients induced by the Wil sonian renormalization group. This yields a direct connection between over parameterization and simplicity of neural network likelihoods. Whether the coefficients are constants or functions may be understood in terms of GP limit symmetries\, as expected from 't Hooft's technical naturalness. Gene ral theoretical calculations are matched to neural network experiments in the simplest class of models allowing the correspondence. Our formalism is valid for any of the many architectures that becomes a GP in an asymptoti c limit\, a property preserved under certain types of training.\n LOCATION:https://researchseminars.org/talk/nhetc-special/1/ END:VEVENT END:VCALENDAR