OoO pipeline / General-Purpose Approximate Programs Hadi Esmaeilzadeh Adrian Sampson University of Washington Luis Ceze / Store Queue Entries Dependence Predictor Caches / Store FUs ROB Entries Issue Queue Entries INT/FP Physical Registers Branch Predictor BTB Sets / FANN library / /
IndustryTerm
software technique / on-line training system / approximate applications / approximate computing literature / approximate computing / potential energy savings / suitable neural network / software implementations / neural networks / energy gains / On-line training / nm technology node / energy-efficient configuration / data mining / trained neural network / image processing algorithms / energy improvements / signal processing / artificial neural networks / hardware neural networks / energy benefits / 3.7× energy improvements / search algorithm / energy reduction / energy savings / energy results / energy / conventional processors / prediction applications / 3.0× energy reduction / remote server / even accounting / energy consumption / neural network / on-line version / application energy reduction / image processing / zero-energy / approximate computing techniques / vision algorithms / gradient descent algorithm / software multiply-and-add / configured neural network / energy-efficient design / software neural network execution / average energy savings / topology search / energy saving / identical processing engines / faster and more energy / diverse applications / energy efficiency / trained neural networks / detection algorithm / backpropagation algorithm / all-software configuration / overall algorithm / search space / software library fft inversek2j jmeint jpeg / topology search algorithm / learning algorithm / applicable image processing computation / /
Organization
University of Washington / Output Register File 8×32-bit Sigmoid Unit LUT 2048×32-bit Multiply-Add Unit / ASIC / /
Person
Pixel / Doug Burger / Washington Luis Ceze / /
Position
speculative head / FPGA Speculative Tail Input FIFO Non-speculative Head Speculative Head / candidate for the Parrot transformation / head / speculative head and a non-speculative head / FIFO Input FIFO Multiply-Add Unit Output Register File Scaling Unit Input FIFO Controller / Imperative Source Code Programmer / representative / scheduler / non-speculative head / programmer / /
Appears in the Proceedings of the 45th Annual IEEE/ACM International Symposium on Microarchitecture, 2012 Neural Acceleration for General-Purpose Approximate Programs Hadi Esmaeilzadeh Adrian Sampson