BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Model-specific DNN Accelerators
DTSTART:20190528T150000
DTEND:20190528T170000
DTSTAMP:20260407T045616Z
UID:efbefbb92f17c9fe0561a9bf7675280dfe9d69570a29322267fb6862
CATEGORIES:Conferences - Seminars
DESCRIPTION:Ahmet Caner Yüzügüler\nEDIC candidacy exam\nExam president:
  Prof. Martin Jaggi\nThesis advisor: Prof. Babak Falsafi\nThesis co-adviso
 r: Prof. Pascal Frossard\nCo-examiner: Prof. Paolo Ienne\n\nAbstract\nThe 
 inefficiencies in the DNN accelerators stem from the generic dataflows\, w
 hich is intended to support large variety of DNN models. However\, the dat
 asets from the same domain usually do not require different DNN model stru
 ctures. For example\, although the datasets for object classification (Ima
 geNet)\,  face recognition (FaceNet) or cancer diagnostics are inherently
  very different from each other\, the models for all these datasets are su
 ccessfully trained with the same DNN structure\, namely GoogleNet. In this
  research\, we propose model-specific DNN accelerators\, which are optimiz
 ed for a single DNN structure. The goal of this research is to study the p
 roperties of DNN structures that make them applicable to the wider range o
 f problems\, and then to show how much computational efficiency one can ga
 in when a hardware accelerator is optimized for this specific DNN structur
 e. \n\nBackground papers\nEfficient Processing of Deep Neural Networks: A
  Tutorial and Survey\, Sections V\, VI and VII from Survey by Sze.\nEyer
 iss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional
 Neural Networks\, by Chen\, Y-H.\, et al.\nSCNN: An Accelerator for Compre
 ssed-sparse Convolutional Neural Networks\, by Parashar\, A.\, et al.\n\n
  
LOCATION:BC 229 https://plan.epfl.ch/?room==BC%20229
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR