BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Date iCal//NONSGML kigkonsult.se iCalcreator 2.20.2//
METHOD:PUBLISH
X-WR-CALNAME;VALUE=TEXT:Eventi DIAG
BEGIN:VTIMEZONE
TZID:Europe/Paris
BEGIN:STANDARD
DTSTART:20181028T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20180325T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.13488.field_data.0@www.diag.uniroma1.it
DTSTAMP:20260406T003845Z
CREATED:20180618T172804Z
DESCRIPTION:Many learning algorithms are struggling with large data sets\, 
 and miss information present in the data simply for computational reasons.
  A larger and mostly hidden problem is that many algorithms learn (uninten
 tionally and unnoticed) triggering patterns that are not supported by the 
 data. Using such classifiers\, we may jump to conclusions that are unjusti
 fiable based on our existing data sets. Both missing important triggers an
 d/or using unsupported ones may result in serious practical/legal problems
 . This brings up both ethical and legal questions. In this talk we demonst
 rate these issues with a small example. We propose the notion of a 'justif
 iable' classifier\, and on the positive side\, we show some results about 
 the existence of learning algorithms that always produce a 'justifiable' c
 lassifier.
DTSTART;TZID=Europe/Paris:20180626T143000
DTEND;TZID=Europe/Paris:20180626T143000
LAST-MODIFIED:20191008T082902Z
LOCATION:Aula Magna DIAG
SUMMARY:SEMINAR - Prof. Endre Boros\, Rutgers University. 'Justifiable and 
 ethical learning - a mathematical view' - Prof. Endre Boros\, Rutgers Univ
 ersity
URL;TYPE=URI:http://www.diag.uniroma1.it/node/13488
END:VEVENT
END:VCALENDAR