BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//UM//UM*Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:America/Detroit
TZURL:http://tzurl.org/zoneinfo/America/Detroit
X-LIC-LOCATION:America/Detroit
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20070311T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20071104T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20261014T095112
DTSTART;TZID=America/Detroit:20261023T150000
DTEND;TZID=America/Detroit:20261023T160000
SUMMARY:Lecture / Discussion:AIM Seminar: How voting goes wrong\, and how it can go right: the evolution of collective intelligence
DESCRIPTION:Abstract: Voting sometimes works well. A crowd's average guess is often better than any individual guess. But voting can also fail\, producing a biased outcome or no decision at all. I will discuss natural mechanisms that cause each of these outcomes\, in groups whose members observe other voters and learn by imitating one another.\n\nI will first describe a voter game\, played online by 2\,520 people\, in which two parties of equal size compete to reach a supermajority. The structure of who sees whose voting intentions\, which we call information gerrymandering\, can bias the outcome toward one party\, even when both parties are the same size and every player has the same influence. When members of both parties see mainly their own party's intentions\, most groups end in deadlock.\n\nI will then ask what individual reward scheme allows a population of self-interested imitators\, each observing only part of a problem\, to evolve accurate collective decisions. For linear prediction problems\, rewarding individuals for their own accuracy eliminates the diversity of opinions a group needs. Rewarding those whose opinions reduce the collective error provably increases accuracy\, until the group predicts perfectly. For more realistic non-linear decisions\, a single majority vote cannot produce accurate decisions under any reward scheme. Voting in layers\, with rewards paid only to individuals who are pivotal at every layer above them\, provably allows a population to evolve accurate decisions. The evolved population is equivalent to a multi-layer perceptron in machine learning\, and the reward for pivotality corresponds to credit assignment that machines compute through backpropagation. These results therefore provide a naturalistic\, evolutionary account of collective in intelligence in humans and machines.\n\nContact: Naoki Masuda
UID:148879-21905041@events.umich.edu
URL:https://events.umich.edu/event/148879
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:Applied Mathematics,Mathematics
LOCATION:East Hall - 1084
CONTACT:
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260807T132324
DTSTART;TZID=America/Detroit:20261023T150000
DTEND;TZID=America/Detroit:20261023T163000
SUMMARY:Workshop / Seminar:Faculty Workshop on Holistic Graduate Admissions for Excellence and Diversity
DESCRIPTION:Holistic admissions practices are an integral part of achieving both diversity and excellence when admitting graduate student cohorts. This workshop is designed to provide concrete strategies and evidence-based practices to support faculty and program goals in admitting diverse and excellent cohorts of graduate students. This workshop will provide members of graduate admissions committees an opportunity to develop their own holistic admissions tools.
UID:150000-21907798@events.umich.edu
URL:https://events.umich.edu/event/150000
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:Rgs Events,Rgs-events,Sessions
LOCATION:Rackham Building, 4th Floor
CONTACT:
END:VEVENT
END:VCALENDAR