BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260402T024533Z
LOCATION:3006\, Level 3
DTSTART;TZID=America/Los_Angeles:20250625T160000
DTEND;TZID=America/Los_Angeles:20250625T161500
UID:dac_DAC 2025_sess158_RESEARCH2570@linklings.com
SUMMARY:MMDFL: Multi-Model-based Decentralized Federated Learning for Reso
 urce-Constrained AIoT Systems
DESCRIPTION:DengKe Yan and Yanxin Yang (East China Normal University), Min
 g Hu (Nanyang Technological University), Xin Fu (University of Houston), a
 nd Mingsong Chen (East China Normal University)\n\nAlong with the prosperi
 ty of Artificial Intelligence (AI) techniques, more and more Artificial In
 telligence of Things (AIoT) applications adopt Federated Learning (FL) to 
 enable collaborative learning without compromising the privacy of devices.
  Since existing centralized FL methods suffer from the problems of single-
 point-of-failure and communication bottleneck caused by the parameter serv
 er, we are witnessing an increasing use of Decentralized Federated Learnin
 g (DFL), which is based on Peer-to-Peer (P2P) communication without using 
 a global model. However, DFL still faces three major challenges, i.e., lim
 ited computing power and network bandwidth of resource-constrained devices
 , non-Independent and Identically Distributed (non-IID) device data, and a
 ll-neighbor-dependent knowledge aggregation operations, all of which great
 ly suppress the learning potential of existing DFL methods. To address the
 se problems, this paper presents an efficient DFL framework named MMDFL ba
 sed on our proposed multi-model-based learning and knowledge aggregation m
 echanism. Specifically, MMDFL adopts multiple traveler models, which perfo
 rm local training individually along their traversed devices, accelerating
  and maximizing knowledge learning and sharing among devices. Moreover, ba
 sed on our proposed device selection strategy, MMDFL enables each traveler
  to adaptively explore its next best neighboring device to further enhance
  the DFL training performance, taking into account issues of data heteroge
 neity, limited resources and catastrophic forgetting phenomenon. Experimen
 tal results from simulation and a real testbed show that, compared with st
 ate-of-the-art DFL methods, MMDFL can not only significantly reduce the co
 mmunication overhead but also achieve better overall classification perfor
 mance for both IID and non-IID scenarios.\n\nTopics: Systems\n\nTracks: SY
 S4: Embedded System Design Tools and Methodologies\n\nSession Chairs: Peip
 ei Zhou (Brown University) and Pi-Cheng Hsiu (Academia Sinica)\n\n
END:VEVENT
END:VCALENDAR