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:20260402T024532Z
LOCATION:3008\, Level 3
DTSTART;TZID=America/Los_Angeles:20250624T160000
DTEND;TZID=America/Los_Angeles:20250624T161500
UID:dac_DAC 2025_sess157_RESEARCH1573@linklings.com
SUMMARY:Enabling On-Tiny-Device Model Personalization via Gradient Condens
 ing and Alternant Partial Update
DESCRIPTION:Zhenge Jia, Yiyang Shi, Zeyu Bao, Zirui Wang, Xin Pang, Huiguo
  Liu, Yu Duan, Zhaoyan Shen, and Mengying Zhao (Shandong University)\n\nOn
 -device training enables the model to adapt to user-specific data by fine-
 tuning a pre-trained model locally. As embedded devices become ubiquitous,
  on-device training is increasingly essential since users can benefit from
  the personalized model without transmitting data and model parameters to 
 the server. Despite significant efforts toward efficient training, on-devi
 ce training still faces two major challenges: (1) the heavy and offline co
 mputation workload required to identify important parameters for updating;
  (2) the prohibitive cost of multi-layer backpropagation, which strains th
 e limited resources of embedded devices. In this paper, we propose an algo
 rithm-system co-optimization framework that enables self-adaptive and on-d
 evice model personalization for resource-constrained embedded devices.  To
  address the challenge of parameter selection, we propose Alternant Partia
 l Update, a method that locally identifies essential parameters without re
 quiring retraining or offline evolutionary search. To mitigate backpropaga
 tion costs, we introduce Gradient Condensing to condense the gradient map 
 structure, significantly reducing the computational complexity and memory 
 consumption of backpropagation while preserving model performance. Our fra
 mework is evaluated through extensive experiments using various CNN models
  (e.g., MobileNet, MCUNet) on embedded devices with minimal resources (e.g
 ., OpenMV-H7 with less than 1MB SRAM and 2MB Flash).  Experimental results
  show that our framework achieves up to 2X speedup, 80% memory saving, and
  30% accuracy improvement on downstream tasks, outperforming SOTA approach
 es.\n\nTopics: Systems\n\nTracks: SYS3: Embedded Software\n\nSession Chair
 s: Hoeseok Yang (Santa Clara University) and Younghyun Kim (Purdue Univers
 ity)\n\n
END:VEVENT
END:VCALENDAR