BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Nonconvex Sparse Regularization For Deep Neural Networks and its O
 ptimal Property
DTSTART;VALUE=DATE-TIME:20191107T083000Z
DTEND;VALUE=DATE-TIME:20191107T084500Z
DTSTAMP;VALUE=DATE-TIME:20260724T053720Z
UID:indico-contribution-18@sshep.snu.ac.kr
DESCRIPTION:Speakers: Ilsang Ohn (Department of Statistics\, Seoul Nationa
 l University)\nSparsity is a key ingredient in the success of learners bot
 h theoretically and computationally. This is also the case for deep neural
  networks (DNNs). A number of empirical observations show that sparse DNNs
  can dramatically reduce computation time and memory without appreciably h
 arming prediction power. Furthermore\, recent theoretical studies proved t
 hat DNN estimators with a certain sparsity constraint can attain optimal c
 onvergence rates for regression and classification problems. However\, the
 y only considered the empirical risk minimizer under the sparsity constrai
 nt\, where optimization is almost impossible in practice due to its discre
 te nature. In this research\, we propose a novel penalized empirical risk 
 minimization method for estimating sparse DNNs with a scalable computation
  algorithm.  The proposed method yields sparse DNNs that can achieve optim
 al convergence rates of excess risks for various learning problems includi
 ng regression and binary classification. We demonstrate the empirical perf
 ormance of the proposed method and compare it with other competitors for v
 arious benchmark datasets.\n\nhttps://sshep.snu.ac.kr/event/107/contributi
 ons/18/
LOCATION:Seoul National University\, College of Natural Sciences\, Buildin
 g 25-1 국제회의실 (International Conference Hall)
URL:https://sshep.snu.ac.kr/event/107/contributions/18/
END:VEVENT
END:VCALENDAR