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:20240626T180033Z
LOCATION:3003\, 3rd Floor
DTSTART;TZID=America/Los_Angeles:20240625T143000
DTEND;TZID=America/Los_Angeles:20240625T144500
UID:dac_DAC 2024_sess158_RESEARCH989@linklings.com
SUMMARY:SpARC: Token Similarity-Aware Sparse Attention Transformer Acceler
 ator via Row-wise Clustering
DESCRIPTION:Research Manuscript\n\nHan Cho, Dongjun Kim, Seungeon Hwang, a
 nd Jongsun Park (Korea University)\n\nIn this paper, we propose SpARC, a s
 parse attention transformer accelerator that enhances throughput and energ
 y efficiency by reducing the computational complexity of the self-attentio
 n mechanism. Our approach exploits inherent row-level redundancies in tran
 sformer attention maps to reduce the overall self-attention computation. B
 y employing row-wise clustering, attention scores are calculated only once
  per cluster to achieve approximate attention without seriously compromisi
 ng accuracy. To leverage the high parallelism of the proposed clustering a
 pproximate attention, we develop a fully pipelined accelerator with a dedi
 cated memory hierarchy.\n\nTopic: AI, Design\n\nKeyword: AI/ML Architectur
 e Design\n\nSession Chair: Hyoukjun Kwon (University of California, Irvine
 )
END:VEVENT
END:VCALENDAR