BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Ro Jefferson (NORDITA) DTSTART:20211109T203000Z DTEND:20211109T214500Z DTSTAMP:20260423T004139Z UID:HET/21 DESCRIPTION:Title: The edge of chaos: quantum field theory and deep neural networks\nby Ro J efferson (NORDITA) as part of Purdue HET\n\n\nAbstract\nThe nascent NN-QFT correspondence offers an interesting\, physics-based approach towards a t heoretical foundation for deep neural networks. In this talk\, I will disc uss recent work in which we explicitly construct the QFT corresponding to a general class of networks encompassing both recurrent and feedforward ar chitectures. The point at which the network exhibits critical behaviour ca n then be obtained by examining the largest Lyapunov exponent in a proper mean field theory treatment. However\, this and previous results in the li terature hold only at infinite width\, while empirical evidence suggests t hat the correlation length receives significant corrections in real-world (finite width) networks. Hence\, going beyond the mean field regime\, we d evelop a perturbative approach formally analogous to that in the O(N) vect or model\, in which the weight variance plays the role of the 't Hooft cou pling. In particular\, we compute both the O(1) corrections quantifying fl uctuations from typicality in the ensemble of networks\, and the subleadin g O(depth/width) corrections due to finite-width effects. I will introduce the use of Feynman diagrams in this context for linear models\, and discu ss the generalization to nonlinear models if time permits.\n LOCATION:https://researchseminars.org/talk/HET/21/ END:VEVENT END:VCALENDAR