BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260409T144021EDT-099824w3lv@132.216.98.100
DTSTAMP:20260409T184021Z
DESCRIPTION:Zhenling Jiang\n\nThe Wharton School\, University of Pennsylvan
 ia\n\nEstimating Treatment Effects under Algorithmic Interference: A Struc
 tured Neural Networks Approach\n\nDate: Friday\, April 10\, 2026\n	Time: 10
 :30 am-12:00 pm EST\n	Location: Bronfman building\, room 310\n\nAll are cor
 dially invited to attend.\n\n\nAbstract:\n\nOnline user-generated content 
 platforms allocate billions of dollars of promotional traffic through algo
 rithms in two-sided marketplaces. To evaluate updates to these algorithms\
 , platforms frequently rely on creator-side randomized experiments. Howeve
 r\, because treated and control creators compete for exposure\, such exper
 iments suffer from algorithmic interference: exposure outcomes depend on c
 ompetitors’ treatment status. We show that commonly used difference-in-mea
 ns estimators can therefore be severely biased and may even recommend depl
 oying inferior algorithms. To address this challenge\, we develop a struct
 ured semiparametric framework that explicitly models the competitive alloc
 ation mechanism underlying exposure. Our approach combines an algorithm ch
 oice model that characterizes how exposure is allocated across competing c
 ontent with a viewer response model that captures engagement conditional o
 n exposure. We construct a debiased estimator grounded in the double machi
 ne learning framework to recover the global treatment effect of platform-w
 ide rollout. Methodologically\, we extend DML asymptotic theory to accommo
 date correlated samples arising from overlapping consideration sets. Using
  Monte Carlo simulations and a large-scale field experiment on a major sho
 rt-video platform\, we show that our estimator closely matches an interfer
 ence-free benchmark obtained from a costly double-sided experimental desig
 n. In contrast\, standard estimators exhibit substantial bias and\, in som
 e cases\, even reverse the sign of the effect.\n
DTSTART:20260410T143000Z
DTEND:20260410T160000Z
LOCATION:Room 310\, Bronfman Building\, CA\, QC\, Montreal\, H3A 1G5\, 1001
  rue Sherbrooke Ouest
SUMMARY:91ºÚÁÏÍø Institute of Marketing (MIM) Seminar: Zhenling Jiang
URL:/desautels/channels/event/mcgill-institute-marketi
 ng-mim-seminar-zhenling-jiang-372235
END:VEVENT
END:VCALENDAR