BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:0
UID:UW-Physics-Event-6485
DTSTART:20210714T160000Z
DTEND:20210714T171500Z
DTSTAMP:20210921T194227Z
LAST-MODIFIED:20210707T195241Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
ysicsmeetsml.org for zoom link
SUMMARY:Learning Differential Equations\, Physics ∩ ML Seminar\, Jes
se Bettencourt\, University of Toronto
DESCRIPTION:Differential equations provide a natural and productive la
nguage to describe and manipulate physical systems. As well\, the inte
rdisciplinary literature developed toward the study of differential eq
uations is rich with conceptual and technical results. I will discuss
the integration of these methods with Machine Learning. I will introdu
ce Neural Ordinary Differential Equations\, a class of initial value p
roblems whose dynamics are specified by a neural network. I will descr
ibe some methods for learning the differential equation via gradient o
ptimization. I will highlight some areas where this treatment is both
conceptually elegant and practically effective. In particular\, I will
discuss Continuous Normalizing Flows for density estimation and an ex
tension (FFJORD) that demonstrates performance improvement through num
erical approximation. I will also discuss recent work to regularize le
arned differential equations such that their solution can be efficient
ly approximated by a numerical solver. I will describe recent advances
in (Higher-Order) Automatic Differentiation that facilitate these met
hods and may be a useful tool for future techniques to study the inter
face of physics and Machine Learning.
URL:https://www.physics.wisc.edu/events/?id=6485
END:VEVENT
END:VCALENDAR