BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20230831T095747Z
LOCATION:Sanada I
DTSTART;TZID=Europe/Stockholm:20230628T150000
DTEND;TZID=Europe/Stockholm:20230628T153000
UID:submissions.pasc-conference.org_PASC23_sess138_msa150@linklings.com
SUMMARY:Designing Application-Specific Approximate Operators for Energy-Ef
 ficient AI Accelerators
DESCRIPTION:Minisymposium\n\nSalim Ullah, Siva Satyendra Sahoo, and Akash 
 Kumar (TU Dresden)\n\nA plethora of recent works has focused on the variou
 s optimization techniques to reduce machine learning (ML) models’ overall 
 computational complexity and memory footprints to implement them on resour
 ce-constrained embedded systems. These techniques mainly exploit the inher
 ent error resilience of ML models to introduce deliberate approximations a
 t the various layers of the computation stack. As Multiply-accumulate (MAC
 ) is the most commonly utilized operation in ML models, most of the relate
 d state-of-the-art works have focused on proposing various approximate arc
 hitectures for multiplication and addition operations. However, most of th
 ese works lack a consistent design methodology. Furthermore, these approxi
 mate arithmetic operators are designed without considering an application’
 s accuracy-performance constraints. The application agnostic-design method
 ology can result in approximate operators, which may not satisfy an applic
 ation’s accuracy-performance constraints. In this session, we will focus o
 n a framework for designing application-specific approximate arithmetic op
 erators for FPGA-based systems. It involves circuit-level modeling (6-inpu
 t Lookup table (LUT) and associated carry chains of modern FPGAs) and nove
 l design space exploration methods to design approximate operators that ca
 n leverage the inherent robustness of ML applications. The framework repor
 ts more non-dominated approximate operators with better hypervolume contri
 bution than state-of-the-art designs for various benchmark applications.\n
 \nDomain: Computer Science, Machine Learning, and Applied Mathematics 
\n\
 nSession Chair: Emma Tolley (EPFL)
END:VEVENT
END:VCALENDAR