BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:2
UID:UW-Physics-Event-6871
DTSTART:20220309T170000Z
DTEND:20220309T181500Z
DTSTAMP:20260414T153808Z
LAST-MODIFIED:20220309T164028Z
LOCATION:Chamberlin 5280 (Zoom link for those attending online: https:
 //uwmadison.zoom.us/j/98994425904?pwd=cnY5REVFaFY5bU1HOUN5V1ZSemdGdz09
  )
SUMMARY:Rethinking AutoML for Diverse Tasks\, Physics ∩ ML Seminar\,
  Nicholas Roberts\, University of Wisconsin-Madison
DESCRIPTION:The underlying motivation of automated machine learning\, 
 or AutoML\, is to automate away tasks which require machine learning (
 ML) expertise--such that experts in domains other than ML can reap its
  potential benefits for their problems. These automation efforts are m
 ostly siloed within the machine learning community by their reliance o
 n the tasks or domains which are most familiar to machine learning exp
 erts--classification tasks in computer vision or NLP. Unfortunately\, 
 this neglects the heavy-tail of tasks and domains that practitioners m
 ight care about and directly contradicts a core premise of AutoML: use
 fulness to non-ML-experts. In this talk\, we present two of our recent
  directions which make progress toward alleviating this issue by expan
 ding into under-explored domains and problem types. In the first half 
 of the talk\, we will present a class of search spaces over deep neura
 l network operations which can specialize a given CNN architecture to 
 any domain of interest by generalizing the convolution theorem from si
 gnal processing. In the second half of the talk\, we will discuss the 
 limitations of weak supervision for semi-automated dataset curation an
 d show how to generalize weak supervision so that it can be applied to
  any label space equipped with a distance metric\, as opposed to categ
 orical labels alone.
URL:https://www.physics.wisc.edu/events/?id=6871
END:VEVENT
END:VCALENDAR