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:Machine Learning @ CUED
SUMMARY:Robust machine learning for causal inference in he
 alth care - David Sontag\, MIT
DTSTART;TZID=Europe/London:20190327T110000
DTEND;TZID=Europe/London:20190327T120000
UID:TALK120847AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/120847
DESCRIPTION:Electronic health records are now pervasive\, pres
 enting an incredible opportunity to use retrospect
 ive data to learn about medicine and to improve he
 alth care. Machine learning can help answer questi
 ons such as\, "What conditions does this patient h
 ave?"\, "When will this patient's disease progress
 ?" and "How should we optimally treat this disease
 ?". Properly answering these questions requires ta
 ckling head-on questions of causality\, specifical
 ly how to infer causality from high-dimensional ob
 servational data. Machine learning and causal infe
 rence in health care introduces additional challen
 ges including little labeled data\, significant mi
 ssing data\, censoring\, and the need to character
 ize individual-level uncertainty. I will discuss s
 everal new methodologies that my group has created
  to address these challenges\, with a particular f
 ocus on disease progression modeling and estimatio
 n of individual treatment effect. Specifically\, I
  discuss provable guarantees for causal inference 
 under model misspecification (Johansson et al. ICM
 L '16\, Shalit et al. ICML '17)\, approaches for c
 ausal inference with unobserved confounding (Louiz
 os et al. NeurIPS '17)\, how to check assumptions 
 for off-policy reinforcement learning (Gottesman e
 t al. Nature Medicine '19\, Oberst et al. '19)\, a
 ssessing overlap (Johansson et al.\, '19)\, and le
 arning nonlinear dynamical models using the deep M
 arkov model (Krishnan et al.\, AAAI '17).
LOCATION:Engineering Department\, CBL Room BE-438.
CONTACT:Adrian Weller
END:VEVENT
END:VCALENDAR