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:20211208T164512Z
LOCATION:3014
DTSTART;TZID=America/Los_Angeles:20211207T153000
DTEND;TZID=America/Los_Angeles:20211207T170000
UID:dac_DAC 2021_sess110@linklings.com
SUMMARY:When They Go High - We Go Low (in Computation)
DESCRIPTION:Research Manuscript\n\nLow computational complexity improves h
 ardware efficiency. To that end, this session reports cutting-edge develop
 ments in the domain of Approximate Computing for AI Hardware-Software code
 sign. Along with novel algorithmic techniques addressing pruning and reduc
 ed precision, the session dives deeper into cross-layer optimization and a
 pproximate arithmetic for different types of deep neural network accelerat
 ors ranging from embedded FPGA systems to in-memory computing.\n\nCOSAIM: 
 Counter-based Stochastic-behaving Approximate Integer Multiplier for Deep 
 Neural Networks\n\nYu\n\nWe propose a new stochastic-behaving approximate 
 integer unsigned multiplier, COSAIM. The new design is an improved stochas
 tic multiplier. Our evaluation shows that the COSAIM with error improvemen
 t can achieve error bias, mean error, peak errors at 0.06%, 0.3%, 1.81%. E
 xperimental results show that...\n\n---------------------\nSoftermax: Hard
 ware/Software Co-Design of an Efficient Softmax for Transformers\n\nSteven
 s\n\nTransformer-based networks are increasingly popular, achieving state-
 of-the-art performance in a number of tasks. This performance is largely a
 ttributed to the use of stacked "self-attention" layers, each of which con
 sists of matrix multiplies as well as softmax operations. As a result, unl
 ike other...\n\n---------------------\nControl Variate Approximation for D
 NN Accelerators\n\nAnagnostopoulos\n\nIn this work, we introduce a control
  variate approximation technique for low error approximate Deep Neural Net
 work (DNN) accelerators. The control variate technique is used in Monte Ca
 rlo methods to achieve variance reduction. Our approach significantly decr
 eases the induced error due to approximate...\n\n\nTopic: Design\n\nKeywor
 d: Approximate Computing for AI/ML\n\nVirtual Program: Presented In-Person
END:VEVENT
END:VCALENDAR