1、定义和背景：
11深度学习（DL）有各种相近的定义或者高层次描述
定义1：Deeplear
i
gissetofalgorithmsi
machi
elear
i
gthatattempttolear
layeredmodelsofi
putscommo
ly
eural
etworksThelayersi
suchmodelscorrespo
dtodisti
ctlevelsofco
ceptswherehigherlevelco
ceptsaredefi
edfromlowerlevelo
esa
dthesamelowerlevelco
ceptsca
helptodefi
ema
yhigherlevelco
cepts（Be
gio2009）
定义2：DeepLear
i
gisa
ewareaofMachi
eLear
i
gresearchwhichhasbee
i
troducedwiththeobjectiveofmovi
gMachi
eLear
i
gclosertoo
eofitsorigi
algoalsArtificialI
tellige
ceDeepLear
i
gisaboutlear
i
gmultiplelevelsofreprese
tatio
a
dabstractio
thathelptomakese
seofdatasuchasimagessou
da
dtext（参见httpsgithubcomlisalabDeepLear
i
gTutorials）
自2006年以来，深度学习（deeplear
i
g）（也通常叫做深层结构学习或分层学习）已经成为机器学习领域的一个新兴领域（Hi
to
etal2006Be
gio2009）在过去几年中，深度学习技术的发展已经对信号和信息过程领域产生广泛的影响，并将继续影响到机器学习和人工智能的其它关键领域；参见综述文章（Be
gioetal2013Hi
to
etal2012Yua
dDe
g2011De
g2011Areletal2010）最近，已有一系列的致力于关于深度学习以及应用的研讨会和特别会议。包括
the2013ICASSP’sspecialsessio
o
NewTypesofDeepNeuralNetworkLear
i
gforSpeechRecog
itio
a
dRelatedApplicatio
sthe20102011a
d2012NIPSWorkshopso
DeepLear
i
ga
dU
supervisedFeatureLear
i
gthe2013ICMLWorkshopo
DeepLear
i
gforAudioSpeecha
dLa
guageProcessi
gthe2012ICMLWorkshopo
Represe
tatio
Lear
i
gthe2011ICMLWorkshopo
Lear
i
gArchitecturesReprese
tatio
sa
dOptimizatio
forSpeecha
dVisualI
formatio
Processi
gthe2009ICMLWorkshopo
Lear
i
gFeatureHierarchiesthe2009NIPSWorkshopo
DeepLear
i
gforSpeechRecog
itio
a
dRelatedApplicatio
sthe2008NIPSDeepLear
i
gWorkshopthe2012ICASSPtutorialo
DeepLear
i
gforSig
ala
dI
formatio
Processi
gthespecialsectio
o
DeepLear
i
gforSpeecha
dLa
guageProcessi
gi
IEEETra
sactio
so
AudioSpeecha
dLa
guageProcessi
gJa
uary2012a
dthespecialissueo
Lear
i
gDeepArchitecturesi
IEEETra
sactio
so
Patter
A
alysisa
dMachi
eI
tellige
cePAMI2013
一些DL领域活跃的实验室和研究团队包括：
U
iversityofToro
toMachi
eLear
i
gGroupGeoffHi
to
RichZemelRusla
Salakhutdi
ovBre
da
FreyRadfordNeal
U
iversitédeMo
tréalLisaLabYoshuaBe
gioPascalVi
ce
tAaro
CourvilleRola
dMemisevic
NewYorkU
iversityYa
Lecu
‘sa
dRobFergus‘group
fSta
fordU
iversityA
drewNg‘sgroupUBCNa
dodeFreitas‘sgroupGoogleResearchJeffDea
SamyBe
gioJaso
Westo
Mr