BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:8fc97c048dff5b114667113e0f875808
DTSTAMP:20261107T123059Z
SUMMARY:Dynamical Systems Seminar – Nicolas Boulle
DESCRIPTION:Speaker: Nicolas Boulle\nTitle: A zero-one law for one-shot sys
 tem identification\nAbstract: Can a model be identified from one experimen
 t? We study analytic systems that are linearly parameterized by a combinat
 ion of prescribed dictionary terms\, such as partial differential operator
 s and dynamical systems. For a single input-response pair\, recovery is po
 ssible exactly when the evaluated dictionary terms are linearly independen
 t. We prove a sharp zero-one law: either no input uniquely determines the 
 coefficients\, or almost every random input sampled from a nondegenerate G
 aussian measure does. This dichotomy reduces one-shot system identificatio
 n to a question about degenerate inputs and provides an a posteriori certi
 ficate for any recovered model. Numerical examples recover dynamical syste
 ms\, nonlinear partial differential equations\, and structured matrix fami
 lies from single trajectory data\, while also detecting when an extra prob
 e is necessary.
URL:https://www.imperial.ac.uk/events/216497/dynamical-systems-seminar-geor
 ge-stepaniants-2/
DTSTART;TZID=Europe/London:20261110T130000
DTEND;TZID=Europe/London:20261110T140000
LOCATION:145\, Huxley Building\, South Kensington Campus\, Imperial College
  London\, London\, SW7 2AZ\, United Kingdom
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:STANDARD
DTSTART:20261110T130000
TZNAME:GMT
TZOFFSETTO:+0000
TZOFFSETFROM:+0000
END:STANDARD
END:VTIMEZONE
END:VCALENDAR
