BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Finite-sample performance of the maximum likelihood estimator in l
 ogistic regression
DTSTART:20260519T091500Z
DTEND:20260519T101500Z
DTSTAMP:20260607T044800Z
UID:indico-event-14482@indico.math.cnrs.fr
DESCRIPTION:Speakers: Jaouad Mourtada (ENSAE)\n\nThe logistic model is a c
 lassical linear model to describe the dependence of binary labels on multi
 variate features. We consider the predictive performance of the maximum li
 kelihood estimator (MLE) for logistic regression\, assessed in terms of lo
 gistic risk. We consider two questions: first\, that of the existence of t
 he MLE—which occurs when the dataset is not linearly separated—\, and 
 second that of its accuracy when it exists. These properties depend on bot
 h the dimension of features and on the signal strength.\nIn the case of Ga
 ussian features and a well-specified logistic model\, we describe sharp qu
 antitative guarantees for the existence and prediction risk of the MLE. We
  then generalize these results in two ways: first\, to non-Gaussian featur
 es satisfying a certain regularity condition\, and second to the case wher
 e the labels no longer follow the logistic model.\n\nhttps://indico.math.c
 nrs.fr/event/14482/
LOCATION:Salle K. Johnson (1R3\, 1er étage)
URL:https://indico.math.cnrs.fr/event/14482/
END:VEVENT
END:VCALENDAR