BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:EDIC mock interview public talk: Towards robust weight sharing neu
 ral architecture search
DTSTART:20201123T160000
DTEND:20201123T170000
DTSTAMP:20260406T194654Z
UID:05dddbdcd0744b0cfb77f35acc37080ccf10a3fe923d55cd5ebdf8db
CATEGORIES:Conferences - Seminars
DESCRIPTION:EDIC PhD candidate\, Kaicheng Yu\, CVLAB\nabstract:\nNeural ar
 chitecture search (NAS) aims to facilitate the design of deep networks for
  new tasks\, and has drawn an increasing attention in the past few years. 
 Weight sharing approach\, that utilizing a super-net to encompass all poss
 ible architectures within the search space\, has become a de facto standar
 d in NAS because it enables the search to be done on commodity hardware. H
 owever\, we find that (i) On average\, some popular NAS algorithms perform
  similarly to the random policy\, (ii) this widely-adopted weight sharing 
 strategy degrades the ranking of the NAS candidates to the point of not re
 flecting their true performance\, thus reducing the effectiveness of the s
 earch process. We further decouple weight sharing from the NAS sampling po
 licy\, and isolates 14 factors of super-net training. To further improve t
 he super-net quality\, we propose a regularization term that aims to maxim
 ize the correlation between the performance rankings of the super-net of t
 he stand-alone architectures using a small set of landmark architectures.\
 n\nshort bio: Kaicheng is a 4th year Ph.D. candidate in the computer visio
 n lab of EPFL\, under the supervision of Dr. Mathieu Salzmann and Prof. Pa
 scal Fua. His research is focusing on improving the neural architecture se
 arch algorithms that automatically discover good architectures with minimu
 m human effort. He is also interested in optimizing convolutional neural n
 etwork (CNN) architectures and operations for mobile devices\, and deploye
 d on fine-grained classification and ego-centric view of hand segmentation
 . Kaicheng is a recipient of the Qualcomm Innovation Fellowship (Europe) 2
 019. He received his Bachelor in Computer Science from the University of H
 ong Kong with first-class honor in 2016.\n\n 
LOCATION:Zoom https://epfl.zoom.us/j/82714899011
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR