BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:America/New_York
X-WR-TIMEZONE:America/New_York
BEGIN:VEVENT
UID:931@fds.yale.edu
DTSTART;TZID=America/New_York:20260209T103000
DTEND;TZID=America/New_York:20260209T113000
DTSTAMP:20260128T214924Z
URL:https://fds.yale.edu/events/fds-seminar-chris-hays-mit-statistics-and-
 strategic-behavior-in-ai-ml-evaluation/
SUMMARY:FDS Seminar: Chris Hays (MIT)\, "Statistics and strategic behavior 
 in AI/ML evaluation"
DESCRIPTION:\nAbstract: Machine learning systems have dramatically reorgani
 zed society\, from what content we consume to whom we hire and how we work
 . To measure their impacts\, we must build trustworthy evaluations. Howeve
 r\, these systems are embedded in complex social environments\, which mean
 s that evaluation data is rarely iid or exogenously determined. Thus\, new
  statistical approaches are needed. In this talk\, I’ll discuss two case
  studies in methods for ML evaluation. In the first\, we analyze generativ
 e AI tournaments\, which rank models based on pairwise human preferences. 
 We’ll focus on standard ranking mechanisms’ non-robustness to strategi
 c candidacy\, where model providers choose which models to submit strategi
 cally in order to maximize the overall performance of their submissions\, 
 and propose a modification which does not suffer from this problem. In the
  second\, we analyze a proposal from the discrimination law literature\, w
 hich would require that ML model developers search for less discriminatory
  hiring\, credit and housing models using randomness in training pipelines
 . We formulate this as a sequential decision-making problem and apply tool
 s from anytime-valid inference to establish when a sufficient search has b
 een conducted.\n\n\n\nBio: Chris Hays is a PhD candidate at the MIT Instit
 ute for Data\, Systems and Society. He is interested in the theory of AI e
 valuations through the lens of statistics and strategic behavior. He is ad
 vised by Manish Raghavan. His PhD is supported by an NDSEG fellowship\, an
 d he has won several awards including a Best Paper award at WWW.\n\n\n\nZo
 om password: 123\n
CATEGORIES:FDS Events,Postdoctoral Applicants
LOCATION:\, \, 
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=\, ;X-APPLE-RADIUS=100;X-TI
 TLE=:geo:0,0
END:VEVENT
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:STANDARD
DTSTART:20251102T010000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR