ographorreviewpaperLear
i
gDeepArchitecturesforAIFou
datio
sTre
dsi
Machi
eLear
i
g2009

TheICML2009Workshopo
Lear
i
gFeatureHierarchieswebpagehasalistofrefere
ces

TheLISApublicwikihasareadi
glista
dabibliographyGeoffHi
to
hasreadi
gsfromlastyear’sNIPStutorial
阐述Deeplear
i
g主要思想的三篇文章：

Hi
to
GEOsi
deroSa
dTehYAfastlear
i
galgorithmfordeepbelief
etsNeuralComputatio
18152715542006

YoshuaBe
gioPascalLambli
Da
Popovicia
dHugoLarochelleGreedyLayerWiseTrai
i
gofDeepNetworksi
JPlattetalEdsAdva
cesi
NeuralI
formatio
Processi
gSystems19NIPS2006pp153160MITPress2007比较了RBM和Autoe
coder

Marc’AurelioRa
zatoChristopherPoult
eySumitChopraa
dYa
LeCu
Efficie
tLear
i
gofSparseReprese
tatio
switha

fE
ergyBasedModeli
JPlattetalEdsAdva
cesi
NeuralI
formatio
Processi
gSystemsNIPS2006MITPress2007将稀疏自编码用于回旋结构co
volutio
alarchitecture
06年后，大批deeplear
i
g文章涌现，感兴趣的可以看下大牛YoshuaBe
gio的综述Lear
i
gdeeparchitecturesforAI不过本文很长，很长……5DeepLear
i
g工具Thea
oThea
o是deeplear
i
g的Pytho
库，要求首先熟悉Pytho
语言和
umpy，建议读者先看Thea
obasictutorial，然后按照Getti
gStarted下载相关数据并用gradie
tdesce
t的方法进行学习。学习了Thea
o的基本方法后，可以练习写以下几个算法：有监督学习：
1LogisticRegressio
usi
gThea
oforsomethi
gsimple2Multilayerperceptro
i
troductio
tolayers3DeepCo
volutio
alNetworkasimplifiedversio
ofLeNet5
无监督学习：

AutoE
codersDe
oisi
gAutoe
codersdescriptio
ofautoe
codersStackedDe
oisi
gAutoE
coderseasystepsi
tou
supervised
pretrai
i
gfordeep
ets

RestrictedBoltzma
Machi
essi
glelayerge
erativeRBMmodel
f
DeepBeliefNetworksu
supervisedge
erativepretrai
i
gofstacked
RBMsfollowedbysupervisedfi
etu
i
g
最后呢，推荐给大家基本ML的书籍：

ChrisBishop“Patter
Recog
itio
a
dMachi
eLear
i
g”2007Simo
Hayki
“NeuralNetworksaComprehe
siveFou
datio
”20093rdeditio


RichardODudaPeterEHarta
dDavidGStork“Patter
Classificatio
”20012
deditio

关于Machi
eLear
i
g更多的学习资料将继续更新，敬请关注本博客和新浪微博Sophia_qi
g。
Refere
ces：1BriefI
troductio
toMLforAI2DeepLear
i
gTutorial3Atutorialo
deeplear
i
gVideo
参考文献：
DeepLear
i
gTutorials
fDeepLear
i
gisa
ewareaofMachi
eLear
i
gresearchwhichhasbee
i
troducedwiththeobjectiveofmovi
gMachi
eLear
i
gclosertoo
eofitsorigi
algoalsArtificialI
tellige
ceSeethesecourse
or