BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Isaac Newton Institute Seminar Series
SUMMARY:Data driven decision making and safety certificate
  synthesis - Kostas Margellos (University of Oxfor
 d)
DTSTART;TZID=Europe/London:20251111T105000
DTEND;TZID=Europe/London:20251111T113000
UID:TALK238462AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/238462
DESCRIPTION:\nData driven algorithms offer a natural framework
  to make decisions in environments affected by unc
 ertainty\, where uncertainty is represented by mea
 ns of data. Neural networks constitute one class o
 f such data driven decision making tools. However\
 , the ``learned&rsquo\;&rsquo\; decisions are inhe
 rently random as they depend on the data used. In 
 this talk we discuss how tools from statistical le
 arning theory based on the notion of compression a
 nd randomized optimization offer a principled fram
 ework to analyze the robustness properties of thes
 e learned decisions. Our results build ``trust on 
 data&rsquo\;&rsquo\;\, and accompany data driven s
 olutions with probabilistic robustness guarantees 
 that capture their generalization properties when 
 it comes to new data\, not included in the learnin
 g/training process. We review recent advancements 
 in this area that allow to boost performance based
  on a data outlier removal procedure. We then show
  how this methodology can be employed to build wha
 t will be referred to as safety informed neural ne
 tworks\, that produce&nbsp\; safety and reachabili
 ty certificates for nonlinear dynamical systems\, 
 accompanying them with prescribed probabilistic gu
 arantees with respect to their validity.\n
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:
END:VEVENT
END:VCALENDAR