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:20260404T111247Z UID:nhetc-special/1 DESCRIPTION:Title: Neural Networks and Quantum Field Theory\nby James Halver son (Northeastern University) as part of Special NHETC Seminar\n\nLecture held in Zoom.\n\nAbstract\nWe propose a theoretical understanding of neura l networks in terms of Wilsonian effective field theory. The correspondenc e relies on the fact that many asymptotic neural networks are drawn from G aussian processes\, the analog of non-interacting field theories. Moving a way from the asymptotic limit yields a non-Gaussian process and correspond s to turning on particle interactions\, allowing for the computation of co rrelation functions of neural network outputs with Feynman diagrams. Minim al non-Gaussian process likelihoods are determined by the most relevant no n-Gaussian terms\, according to the flow in their coefficients induced by the Wilsonian renormalization group. This yields a direct connection betwe en overparameterization and simplicity of neural network likelihoods. Whet her the coefficients are constants or functions may be understood in terms of GP limit symmetries\, as expected from 't Hooft's technical naturalnes s. General theoretical calculations are matched to neural network experime nts in the simplest class of models allowing the correspondence. Our forma lism is valid for any of the many architectures that becomes a GP in an as ymptotic limit\, a property preserved under certain types of training.\n LOCATION:https://stable.researchseminars.org/talk/nhetc-special/1/ END:VEVENT END:VCALENDAR