共4个m文件第一个m文件r
命名为DeepLear
i
gmr
r
功能：演示深度学习算法在计算机视觉中的应用r
训练DBN用于分类；r
环境：Wi
7，Matlab2012br
ModiNUDTVAPr
时间：20141010r
r
r
fu
ctio
DeepLear
i
gr
r
loadtrai
i
gdatar
clearr
dataloaddatam
istmatr
trai
_xdoubledatatrai
_x255r
trai
_ydoubledatatrai
_yr
test_xdoubledatatest_x255r
test_ydoubledatatest_yr
cleardatar
r
setupDBNmodelr
i
put_dimsizetrai
_x2r
output_dimsizetrai
_y2r
hidde
_sz1100r
hidde
_sz2100r
db
sizesi
put_dimhidde
_sz1hidde
_sz2r
db
_opts
umepochs50r
db
_optsbatchsize100r
db
_optsmome
tum05r
db
_optsalpha1r
fori1
umeldb
sizes1r
db
rbmialphadb
_optsalphar
db
rbmimome
tumdb
_optsmome
tumr
db
rbmiWzerosdb
sizesi1db
sizesir
db
rbmivWzerosdb
sizesi1db
sizesir
db
rbmibzerosdb
sizesi1r
db
rbmivbzerosdb
sizesi1r
db
rbmiczerosdb
sizesi11r
db
rbmivczerosdb
sizesi11r
e
dr
r
layerwisepretrai
DBNmodelr
xtrai
_xr
r
gdefaultr
db
rbm1rbmtrai
db
rbm1xdb
_optsr
fori2
umeldb
sizes1r
xsigmrepmatdb
rbmi1csizex11xdb
rbmi1Wr
db
rbmirbmtrai
db
rbmixdb
_optsr
e
dr
r
useDBNtoi
itializeNNtopeformfi
etu
i
gr

sizesdb
sizesoutput_dimr


umel
sizesr

lear
i
g_rate01r

mome
tum05r
r

_opts
umepochs50r

_optsbatchsize100r
r
r
r
gdefaultr
fori1
umeldb
rbmr

Widb
rbmicdb
rbmiWr

vWizerossize
Wir
e
dr
i

1r

Wira
d
sizesi1
sizesi10524sqrt6
sizesi1
sizesir

vWizerossize
Wir
r


trai

trai
_xtrai
_y
_optsr
err_rate
test
test_xtest_yr
r
With50epochstheerrorratecouldbearou
d5r
dispFi
alclassificatio
errorrate
um2strerr_rate100r
r
e
dr
r
fu
ctio
XsigmPr
X11expPr
e
dr
r
r
r
第二个m文件命名为
testmr
fu
ctio
err_rateerr_
um
test
xyr


ff
xzerossizex1
sizese
dr
labelsmax
ae
d2r
expectedmaxy2r
err_
umfi
dlabelsexpectedr
err_rate
umelerr_
umsizex1r
e
dr
r
fu
ctio


ff
xyr


r
msizex1r
xo
esm1xr

a1xr
feedforwardpassr
fori2
1r

aisigm
ai1
Wi1r
Addthebiastermr

aio
esm1
air
e
dr

a
r