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:20211208T164407Z
LOCATION:3014
DTSTART;TZID=America/Los_Angeles:20211207T133000
DTEND;TZID=America/Los_Angeles:20211207T135300
UID:dac_DAC 2021_sess141_RESEARCH1098@linklings.com
SUMMARY:MAT: Processing In-Memory Acceleration for Long-Sequence Attention
DESCRIPTION:Research Manuscript\n\nMAT: Processing In-Memory Acceleration 
 for Long-Sequence Attention\n\nZhou\n\nProcessing attention-based machine 
 learning models can be prohibitively costly on long sequences because of t
 he large memory consumption. In this work, we propose MAT, a processing in
 -memory framework, to accelerate long-sequence attention. MAT adopts a mem
 ory-efficient processing flow for attention models that can process sub-se
 quences in a pipeline with a much smaller memory footprint. MAT utilizes t
 wo techniques, reuse-driven data layout, and optimal sample scheduling, to
  optimize the performance of memory-efficient attention. Our experiments s
 how that MAT provides significant speedups and energy efficiency improveme
 nts over TPU and GPU on two emerging long-sequence tasks, medical image pr
 ocessing, and natural language processing.\n\nTopic: Embedded Systems\n\nK
 eyword: Near-Memory and In-Memory Computing\n\nVirtual Program: Presented 
 In-Person
END:VEVENT
END:VCALENDAR