BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Stabilizing black-box model selection
DTSTART:20250707T090000Z
DTEND:20250707T093000Z
DTSTAMP:20260610T221900Z
UID:indico-event-14491@indico.math.cnrs.fr
DESCRIPTION:Speakers: Rebecca Willett\n\nModel selection is the process of
  choosing from a class of candidate models given data. For instance\, we m
 ay wish to select which set of features best predict a label or response o
 r select an equation that hypothesizes a model of a dynamic biological pro
 cess. However\, absent strong assumptions\, typical approaches to these pr
 oblems are highly unstable: if a single data point is removed from the tra
 ining set\, a different model may be selected. In this talk\, I will prese
 nt a new approach to stabilizing model selection with theoretical stabilit
 y guarantees that leverages a combination of bagging and an ''inflated'' a
 rgmax operation. Our method selects a small collection of models that all 
 fit the data\, and it is stable in that\, with high probability\, the remo
 val of any training point will result in a collection of selected models t
 hat overlap with the original collection. We illustrate this method in a m
 odel selection problem focused on identifying how competition in an ecosys
 tem influences species' abundances and a graph estimation problem using ce
 ll-signaling data from proteomics. In these settings\, the proposed method
  yields stable\, compact\, and accurate collections of selected models\, o
 utperforming a variety of benchmarks. This is joint work with Melissa Adri
 an and Jake Soloff. \n\nhttps://indico.math.cnrs.fr/event/14491/
URL:https://indico.math.cnrs.fr/event/14491/
END:VEVENT
END:VCALENDAR