BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Econometrics Workshop:  Nathan Kallus
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260908T001652Z
UID:tag:localist.com\,2008:EventInstance_46305598992201
DTSTART:20240507T154000Z
DTEND:20240507T165500Z
DESCRIPTION:Nathan Kallus\, Cornell Tech\n\nDebiased Inference on Functiona
 ls of Inverse Problems and Applications to Long-Term Causal Inference\n\nI
 n the presence of endogeneity\, instruments and negative controls can stil
 l give us a view onto causal effects\, but only indirectly\, e.g.\, as a f
 unction whose residuals are orthogonal to instruments. Without imposing (u
 nrealistic) parametric restrictions\, these inverse problems are generally
  ill posed\, making it difficult to reliably learn a solution from data. I
 n this talk I discuss how to nonetheless make reliable inferences on linea
 r functionals of (nonparametric) solutions to these inverse problems\, suc
 h as average effects. Any such parameter admits a doubly robust representa
 tion involving the solution to a dual inverse problem that is specific to 
 the functional of interest. We use this to develop debiased estimators tha
 t are root-n-asymptotically normal around the parameter as long as either 
 the primal or dual inverse problem is sufficiently well posed compared to 
 the functional complexity of the (generic\, nonparametric) hypothesis clas
 ses for the solutions to the inverse problems\, all without knowledge of w
 hich inverse problem is the more well posed or how well posed. The result 
 is enabled by strong guarantees for a new iterated Tikhonov regularized ad
 versarial learner for solutions to inverse problems over general hypothesi
 s classes. I will then discuss the particular problem of using the plethor
 a of A/B tests undertaken on digital platforms for learning better surroga
 te indices for inference on long-term causal effects from short-term exper
 iments. While this too can be phrased as a functional of an instrumental v
 ariable regression\, since each A/B test has a fixed size\, here we encoun
 ter the additional challenge of weak instruments\, introducing a non-vanis
 hing bias. We resolve this by learning the nuisances for our debiased esti
 mator using a jackknifed loss function that eliminates this bias and recov
 ers consistency if we have many\, albeit weak\, instruments.
GEO:42.447296;-76.482254
LOCATION:Uris Hall\, 498
SUMMARY:Econometrics Workshop:  Nathan Kallus
URL;VALUE=URI:https://events.cornell.edu/event/econometrics-workshop-nathan
 -kallus
CATEGORIES:Seminar
CATEGORIES:Class/ Workshop
END:VEVENT
END:VCALENDAR