BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Reasoning Under Uncertainty: Trustworthy LLMs for High-Stakes Medi
 cine
DTSTART:20260626T093000
DTEND:20260626T113000
DTSTAMP:20260510T081803Z
UID:1fcf62a1d9cccd53d633e091a520cb9195ffb3768181f01965c154ff
CATEGORIES:Conferences - Seminars
DESCRIPTION:Yusuf Kesmen\nEDIC candidacy exam\nExam president: Prof. Rober
 t West\nThesis advisor: Prof. Mary-Anne Hartley\nCo-examiner: Prof. Martin
  Jaggi\n\nAbstract\nCurrent machine learning systems are very good at task
 s reducible to statistical pattern recognition but falter on problems requ
 iring deliberate\, structured inference. Clinical reasoning exemplifies th
 e latter: physicians generate hypotheses abductively\, revise beliefs prob
 abilistically as evidence accumulates\, and choose among actions whose cos
 ts are deeply asymmetric. The harm of a missed diagnosis is not the harm o
 f an unnecessary test. Large language models\, increasingly deployed for c
 linical decision support\, inherit this limitation. They conflate knowledg
 e with inference: the same parameters that store medical facts are asked t
 o reason over them\, which leaves no explicit mechanism to quantify uncert
 ainty\, to abstain when evidence is insufficient\, or to weight outcomes b
 y their true clinical costs. This work argues that progress toward trustwo
 rthy medical AI requires separating the representation of knowledge from t
 he machinery of inference\, and grounding the latter in formal frameworks 
 drawn from statistical decision theory. We investigate how these framework
 s can be embedded into LLM reasoning pipelines and imposed as external sca
 ffolding around them\, maintaining competing hypotheses\, guiding evidence
  gathering\, and selecting actions that reflect the asymmetric cost struct
 ure of clinical practice. The goal is a class of systems capable of reason
 ing over their uncertainty\, systems that know when they do not know.\n\nS
 elected papers\ncoming soon
LOCATION:INF 326 https://plan.epfl.ch/?room==INF%20326
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR