[Theory] NOW: [TTIC Talks] 10/3 TTIC Colloquium: John C. Duchi, Stanford University

[Brandie Jones]

*When:*        Monday, October 3rd at *11:30am CT*

*Where:      *Talk will be given *live, in-person* at

                      TTIC, 6045 S. Kenwood Avenue

                      5th Floor, Room 530

*Virtually: *via Panopto (Join Here
<https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=d75fa37a-c24c-4dda-a193-af1901495c36>
)

*Who:          *John C. Duchi, Stanford University

*Title:*          How many labelers do you have? Some perspectives on "gold
standard" labels

*Abstract: *The construction of most supervised learning datasets revolves
around collecting multiple labels for each instance, then aggregating the
labels to form a type of “gold-standard” label. We question the wisdom of
this pipeline by developing a (stylized) model of this process and
analyzing its statistical consequences, showing how access to
non-aggregated label information can make training well-calibrated models
easier or—in some cases—even feasible, whereas it is impossible with only
gold-standard labels. The entire story, however, is subtle, and the
contrasts between aggregated and fuller label information depend on the
particulars of the problem. The theory we develop in the stylized model
makes several predictions for real-world datasets, including when
non-aggregate labels should improve learning performance, which we
carefully test to corroborate the validity of our predictions.

Based on joint work with Chen Cheng and Hilal Asi

*Host: Nathan Srebro <nati at ttic.edu>*



-- 
*Brandie Jones *
*Administrative Assistant*
Toyota Technological Institute
6045 S. Kenwood Avenue
Chicago, IL  60637
www.ttic.edu

Working Remote on Tuesdays
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20221003/1644d8e4/attachment.html>