BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Seminar by Jean Pauphilet\, MIT
DTSTART:20190905T150000
DTEND:20190905T163000
DTSTAMP:20260506T144841Z
UID:9356b7c2217095ead6ca0588e773a1f0c63f9c5a568e0a2c15ef0f57
CATEGORIES:Conferences - Seminars
DESCRIPTION:Jean Pauphilet\, MIT\nInterpretable Analytics for Healthcare O
 perations: New algorithms and implementation \n\nAbstract\n\nThe majority 
 of hospitals in the US are experiencing significant operational and financ
 ial stress due to limited capacity\, escalating costs and waning reimburse
 ments. Analytics could certainly alleviate this burden and improve healthc
 are delivery\, provided that they answer industry concerns\, chief among w
 hich is the issue of interpretability. In this talk\, we address these cha
 llenges by designing and implementing new interpretable machine learning t
 echniques. \nIn the first part of the talk\, we develop a machine learnin
 g framework to predict inpatient flows\, namely imminent discharges\, long
 -stay patients\, discharge destination and need for intensive care. Our me
 thods are now fully implemented into the Electronic Health Record system o
 f a major hospital in Boston and provide daily predictions with 80% accura
 cy\, alongside patient-specific explanations. We obtain interpretable insi
 ghts from using decision tree models and advanced feature selection techni
 ques.\nIn the second part of the talk\, we present in more detail the nove
 l algorithmic methods we used for feature selection. We consider the probl
 em of training a linear predictive model under sparsity constraint. We pro
 pose a combination of a first-order heuristic and a cutting-plane algorith
 m to successfully compute high-quality solutions for problems with up to 1
 00\,000s covariates. On extensive numerical experiments\, we achieve more 
 accurate feature selection than competitors\, with comparable predictive p
 ower. \n\n 
LOCATION:ODY 4 03 https://plan.epfl.ch/?room==ODY%204%2003
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR