BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000713Z
LOCATION:704-706
DTSTART;TZID=America/Denver:20231112T165800
DTEND;TZID=America/Denver:20231112T171800
UID:submissions.supercomputing.org_SC23_sess437_ws_shda102@linklings.com
SUMMARY:Pareto Optimization of CNN Models via Hardware-Aware Neural Archit
 ecture Search for Drainage Crossing Classification on Resource-Limited Dev
 ices
DESCRIPTION:Yuke Li (University of California, Merced); Jiwon Baik (Univer
 sity of California, Santa Barbara); Md Marufi Rahman (University of North 
 Texas); Iraklis Anagnostopoulos and Ruopu Li (Southern Illinois University
 ); and Tong Shu (University of North Texas)\n\nEmbedded devices, constrain
 ed by limited memory and processors, require deep learning models to be ta
 ilored to their specifications. This research explores customized model ar
 chitectures for classifying drainage crossing images. Building on the foun
 dational ResNet-18, this paper aims to maximize prediction accuracy, reduc
 e memory size, and minimize inference latency. Various configurations were
  systematically probed by leveraging hardware-aware neural architecture se
 arch, accumulating 1,717 experimental results over six benchmarking varian
 ts. The experimental data analysis, enhanced by nn-Meter, provided a compr
 ehensive understanding of inference latency across four different predicto
 rs. Significantly, a Pareto front analysis with three objectives of accura
 cy, latency, and memory resulted in five non-dominated solutions. These st
 andout models showcased efficiency while retaining accuracy, offering a co
 mpelling alternative to the conventional ResNet-18 when deployed in resour
 ce-constrained environments. The presentation concludes by highlighting in
 sights drawn from the results and suggesting avenues for future exploratio
 n.\n\nTag: Accelerators, Codesign, Heterogeneous Computing, Task Paralleli
 sm\n\nRegistration Category: Workshop Reg Pass\n\nSession Chairs: Tong Shu
  (University of North Texas), Seung-Hwan Lim (Oak Ridge National Laborator
 y (ORNL)), Pavan Balaji (Meta AI), and Sanjukta Bhowmick (University of No
 rth Texas)\n\n
END:VEVENT
END:VCALENDAR