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:20260402T024534Z
LOCATION:3001\, Level 3
DTSTART;TZID=America/Los_Angeles:20250624T111500
DTEND;TZID=America/Los_Angeles:20250624T113000
UID:dac_DAC 2025_sess102_RESEARCH2291@linklings.com
SUMMARY:Mixed-Precision Quantization for Deep Vision Models with Integer Q
 uadratic Programming
DESCRIPTION:Zihao Deng (The University of Texas at Austin), Sayeh Sharify 
 and Xin Wang (d-Matrix Corp.), and Michael Orshansky (The University of Te
 xas at Austin)\n\nQuantization is a widely used technique to compress neur
 al networks. Assigning uniform bit-widths across all layers can result in 
 significant accuracy degradation at low precision and inefficiency at high
  precision. Mixed-precision quantization (MPQ) addresses this by assigning
  varied bit-widths to layers, optimizing the accuracy-efficiency trade-off
 . Existing sensitivity-based methods for MPQ assume that quantization erro
 rs across layers are independent, which leads to suboptimal choices. We in
 troduce CLADO, a practical sensitivity-based MPQ algorithm that captures c
 ross-layer dependency of quantization error. CLADO approximates pairwise c
 ross-layer errors using linear equations on a small data subset. Layerwise
  bit-widths are assigned by optimizing a new MPQ formulation based on cros
 s-layer quantization errors using an Integer Quadratic Program. Experiment
 s with CNN and transformer models on ImageNet demonstrate that CLADO achie
 ves state-of-the-art mixed-precision quantization performance. Code is ava
 ilable at \href{https://anonymous.4open.science/r/CLADO_DAC2025-BC55/READM
 E.md}{CLADO anonymous GitHub repo.}\n\nTopics: AI\n\nTracks: AI1: AI/ML Al
 gorithms\n\nSession Chairs: Chenhui Deng (Nvidia) and Liu Cheng (Chinese A
 cademy of Sciences, University of Chinese Academy of Sciences)\n\n
END:VEVENT
END:VCALENDAR