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:20260730T152640Z
LOCATION:Exhibit Hall
DTSTART;TZID=America/Los_Angeles:20260728T173700
DTEND;TZID=America/Los_Angeles:20260728T173700
UID:dac_DAC 2026_sess306_LBR032@linklings.com
SUMMARY:Late Breaking Results: Efficient MoE-ViT Gating on Edge Accelerato
 r via Dynamic Expert Routing and Quantized Router
DESCRIPTION:Zheqi He, Zhendong Zheng, and teng wang (University of Science
  and Technology of China); Nanhao Zhou (13915576921); Lei Gong (University
  of Science and Technology of China); Wenqi Lou (USTC); Chao Wang (Univers
 ity of Science and Technology of China); and Xuehai Zhou (USTC)\n\nWe prop
 ose a dynamic routing strategy that adapts expert usage to input difficult
 y while reducing overall computation. We also quantize the gating weights 
 using a multiplier-free scheme and apply\npost-training calibration to pre
 serve routing decisions. Compared with Top-K routing, dynamic routing impr
 oves multi-task performance by 5.07%. During inference on the ZCU102, it r
 educes MoE\nexpert FFN computation by 24.71% and lowers end-to-end latency
  by 4.3%, while gating quantization reduces DSP usage by 85%.\n\nTrack: St
 udent\n\n
END:VEVENT
END:VCALENDAR