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:20260730T152638Z
LOCATION:
DTSTART;TZID=America/Los_Angeles:20260728T125600
DTEND;TZID=America/Los_Angeles:20260728T125700
UID:dac_DAC 2026_sess315_RESEARCH1999@linklings.com
SUMMARY:HAP: Efficient Quantization Harnessing Adaptive Precision for DNN 
 Hardware Acceleration
DESCRIPTION:Erjing Luo (University of Alberta), Xinkuang Geng and Honglan 
 Jiang (Shanghai Jiao Tong University), leibo liu (Tsinghua University), an
 d Jie Han (University of Alberta)\n\nPost-training quantization (PTQ) avoi
 ds expensive training process and it is effective in accelerating Deep Neu
 ral Networks (DNNs). Since the representation capacity of a quantizer decr
 eases exponentially with precision, PTQ encounters fundamental limitations
  in the low-precision domain. As an alternative, adaptive-precision quanti
 zation (APQ) leverages the sparsity in data distribution to reduce precisi
 on without sacrificing resolution. This is especially suitable for activat
 ions, as they tend to be sparser and more vulnerable than weights. However
 , existing APQ approaches require data to heavily cluster near zeros, whic
 h is not guaranteed in asymmetric quantization that is preferred for activ
 ations. Furthermore, the varying precision can lead to an unbalanced compu
 tational workload, making it difficult to effectively harvest the theoreti
 cal performance gain. To solve these problems, we propose a novel quantiza
 tion and accelerator design, harnessing adaptive-precision (HAP). It lever
 ages a per-group dynamic zero-point to generalize APQ. Moreover, its intra
 -channel grouping strategy makes it possible to balance variable-precision
  workload via reordering. Leveraging a channel-level dual-precision weight
  quantization scheme, it achieves superior accuracy compared with existing
  PTQ solutions at the level of 4 or 5 bits for a variety of DNN families. 
 On hardware, a novel bit-serial accelerator featuring a lightweight reorde
 r engine is developed. Results show it achieves a 2.65x speedup and a 55% 
 energy reduction on average compared to existing accelerators.\n\nTopics: 
 Quantum\n\nTrack: DES6. Quantum Computing, Systems\n\nSession Chairs: Hoke
 un Kim (Arizona State University); BaekGyu Kim (Daegu Gyeongbuk Institute 
 of Science and Technology); Xiang Chen (Peking University); Sangyeob Kim (
 Yonsei University); Robert M. Radway (University of Pennsylvania); Prashan
 th Mohan (University of Pittsburgh); Xiaoxuan Yang (University of Virginia
 , Stanford University); Ramtin Zand (University of South Carolina); Hadi K
 amali (University of Central Florida); Jiafeng Xie (Villanova University);
  Prabuddha Chakraborty (University of Maine); Junichiro Kadomoto (Universi
 ty of Tokyo); and Yosuke Ueno (RIKEN)\n\n
END:VEVENT
END:VCALENDAR