大数据分析的无限深度神经网络方法.pdf
文本预览下载声明
计算机研究与发展
of and
Journal Research 53(1):68—79,2016
Computer Development
大数据分析的无限深度神经网络方法
张 蕾 章 毅
(四川大学计算机学院机器智能实验室成都610065)
(1eizhang@scu.edu.cn)
Data InfiniteNeuralNetworks
Big Analysisby Deep
Leiand Yi
Zhang Zhang
(^,妃f^i订P,咒z已zzigP卵f已Ln60r口fory,CoZZPgPo,CoⅢ户“£PrSfiP超fP,Sif丘“n行L7”i口Prsi£了,C^P以gd“610065)
Abstractneuralnetworks(DNNs)andtheir arewellknownintheacademic
Deep learningalgorithms
and asthemostsuccessfulmethodsfor data with
communityindustry big analysis.Compared
traditional methodsusedata—drivenandcanextractfeatures
methods,
deeplearning (knowledge)
from methodshave in
automaticallydata.[)eep1earning significantadvantagesanalyzing
andvariedmodelandcrossfield data.At most used
unstructured,unknown big widely
present,the
neuralnetworksin data arefeedforwardneural workwell
deep big analysis networks(FNNs).They
in thecorrelationfromstaticdataand fordata scenariosbasedon
extracting suiting application
Butlimiteditsintrinsic the neuralnetworksto
classification. structurc, offccdforward
by ability
extracttime featuresisweak.Infiniteneuralnetworks,i.e.recurrentneuralnetworks
sequence deep
(RNNs)are Theiressentialcharacteristhatthestatesofthenetworks
dynamicalsystemsessentially.
withtimeand thetime Henceare
显示全部