BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Algorithmic stability for regression and classification - Rina Foy
 gel Barber  (Chicago)
DTSTART:20250313T163000Z
DTEND:20250313T173000Z
UID:TALK219178@talks.cam.ac.uk
CONTACT:HoD Secretary\, DPMMS
DESCRIPTION:In a supervised learning setting\, a model fitting algorithm i
 s unstable if small perturbations to the input (the training data) can oft
 en lead to large perturbations in the output (say\, predictions returned b
 y the fitted model). Algorithmic stability is a desirable property with ma
 ny important implications such as generalization and robustness\, but test
 ing the stability property empirically is known to be impossible in the se
 tting of complex black-box models. In this work\, we establish that baggin
 g any black-box regression algorithm automatically ensures that stability 
 holds\, with no assumptions on the algorithm or the data. Furthermore\, we
  construct a new framework for defining stability in the context of classi
 fication\, and show that using bagging to estimate our uncertainty about t
 he output label will again allow stability guarantees for any black-box mo
 del. This work is joint with Jake Soloff and Rebecca Willett.\n\n\n\nA win
 e reception in the Central Core will follow this lecture
LOCATION:Centre for Mathematical Sciences MR2
END:VEVENT
END:VCALENDAR