圖3.9計(jì)算性能對比
圖3.10歸一化單位成本對比
圖3.11為實(shí)際業(yè)務(wù)中利用FPGA進(jìn)行加速的情況，由圖中數(shù)據(jù)可知FPGA加速可以有效降低成本。
圖3.11某實(shí)際業(yè)務(wù)中的性能和成本對比
參考文獻(xiàn)
[1]AlexKrizhevsky.ImageNetClassificationwithDeepConvolutionalNeuralNetworks
[2]C.Zhang,etal.OptimizingFPGA-basedacceleratordesignfordeepconvolutionalneuralnetworks.InACMISFPGA2015.
[3]PGysel,MMotamedi,SGhiasi.Hardware-orientedApproximationofConvolutionalNeuralNetworks.2016.
[4]SongHan,HuiziMao,WilliamJ.Dally.DEEPCOMPRESSION:COMPRESSINGDEEPNEURALNETWORKSWITHPRUNING,TRAINEDQUANTIZATIONANDHUFFMANCODING.ConferencepaperatICLR,2016
登陸|注冊歡迎登陸本站，認(rèn)識更多朋友，獲得更多精彩內(nèi)容推薦！