BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Deep Harmony: Balancing Utilization and Generality in the Era of N
 eural Innovation
DTSTART:20190130T100000
DTEND:20190130T110000
DTSTAMP:20260407T055553Z
UID:15c918acec33943c40423b56b5df42bc212fb5a24a2f5b6d4791f1d4
CATEGORIES:Conferences - Seminars
DESCRIPTION:Hadi Esmaeilzadeh\, Professor at University of California\, Sa
 n Diego\, USA\nAbstract: Generative Adversarial Networks (GANs) are one of
  the most recent deep learning models that generate synthetic data from li
 mited genuine datasets. GANs are on the frontier as further extension of d
 eep learning into many domains (e.g.\, medicine\, robotics\, content synth
 esis) requires massive sets of labeled data that is generally either unava
 ilable or prohibitively costly to collect. Although GANs are gaining promi
 nence in various fields\, there were no accelerators for these new models.
  In fact\, GANs leverage a new operator\, called transposed convolution\, 
 that exposes unique challenges for hardware acceleration. This operator fi
 rst inserts zeros within the multidimensional input\, then convolves a ker
 nel over this expanded array to add information to the embedded zeros. Eve
 n though there is a convolution stage in this operator\, the inserted zero
 s lead to underutilization of the compute resources when a conventional co
 nvolution accelerator is employed. This talk presents the GANAX architectu
 re to alleviate the sources of inefficiency associated with the accelerati
 on of GANs using conventional convolution accelerators\, making the first 
 GAN accelerator design possible. We propose a reorganization of the output
  computations to allocate compute rows with similar patterns of zeros to a
 djacent processing engines\, which also avoids inconsequential multiply-ad
 ds on the zeros. The talk then focuses on how to reduce underutilization u
 sing runtime information\, Specifically\, in numerous modern deep model\, 
 the outputs of compute-heavy convolution operations are fed to activation 
 units that output zero if their input is negative. By exploiting this uniq
 ue algorithmic property\, we propose a predictive early activation techniq
 ue\, dubbed SnaPEA. This technique cuts the computation of convolution ope
 rations short if it determines that the output will be negative. SnaPEA ca
 n operate in two distinct modes\, exact and predictive. In the exact mode\
 , with no loss in classification accuracy\, SnaPEA statically re-orders th
 e weights based on their signs and periodically performs a single-bit sign
  check on the partial sum. The encouraging results from these innovations 
 how balanced designs pave the way for accelerating the next generation of 
 deep neural models.\n\nBio: Dr. Esmaeilzadeh was awarded early tenure at t
 he University of California\, San Diego (UCSD)\, where he is the inaugural
  holder of Halicioglu Chair in Computer Architecture with the rank of asso
 ciate professor in Computer Science and Engineering. Prior to UCSD\, he wa
 s an assistant professor in the School of Computer Science at the Georgia 
 Institute of Technology from 2013 to 2017. There\, he was the inaugural ho
 lder of the Catherine M. and James E. Allchin Early Career Professorship. 
 Hadi is the founding director of the Alternative Computing Technologies (A
 CT) Lab\, where his team is developing new technologies and cross-stack so
 lutions to build the next generation computer systems. He is also the asso
 ciate director of Center for Machine Integrated Compu=ng and Security (MIC
 S) at UCSD. Dr. Esmaeilzadeh obtained his Ph.D. from the Department of Com
 puter Science and Engineering at the University of Washington in 2013 wher
 e his Ph.D. work received the 2013 William Chan Memorial Best Dissertation
  Award. Prof. Esmaeilzadeh received the IEEE Technical Committee on Comput
 er Architecture (TCCA) “Young Architect” Award in 2018 and was inducte
 d to the ISCA Hall of Fame in the same year. He has received the Air Force
  Office of Scientific Research Young Investigator Award (2017)\, College o
 f Computing Outstanding Junior Faculty Research Award (2017)\, Qualcomm Re
 search Award (2017 and 2016)\, Google Research Faculty Award (2016 and 201
 4)\, Microsoft Research Award (2017 and 2016)\, and Lockheed Inspirational
  Young Faculty Award (2016). His teams were awarded the Qualcomm Innovatio
 n Fellowship in 2014 and 2018\, one of his students was a Microsoft Resear
 ch Fellow\, and another won the 2017 National Center for Women & IT (NCWIT
 ) Collegiate Award. Four of his undergraduate students have been awarded t
 he Georgia Tech President’s Undergraduate Research Award (PURA). His res
 earch has been recognized by four Communications of the ACM Research Highl
 ights\, four IEEE Micro Top Picks\, a nomination for Communications of the
  ACM Research Highlights\, an honorable mention in IEEE Micro Top Picks\, 
 and a Distinguished Paper Award in HPCA 2016. Hadi’s work on dark silico
 n has also been profiled in New York Times. More information is available 
 on his webpage\, http://cseweb.ucsd.edu/~hadi/.\n 
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR