際際滷

際際滷Share a Scribd company logo
Dr gafar zen alabdeen salh
(2011) 1
悵悋惠悸 悋惺惶惡悸 悋愆惡悋惠悋惠惴(2)
self-organizing neural network
悒惺惆悋惆:惆.惶悋忰 悋惺悋惡惆 慍 悴惺惘
惠悋悸 悋忰悋愕惡 惺 悸 悋 悴悋惺悸
悋惺悋惠
悋惠悋愕 悋惠惺competitive
learning
悋惺悋愆悋悧惺悋悖悽惘惠惺惆悒愆惘悋悋惠惺
悋惠悋愕忰惓惠惠悋愕悋惺惶惡悋惠惺惡惺惷悋悋惡惺惷
惺悋惠愆愀.悖惓悋悄悋惠惺悋惡悋惠愆愀惺惆惆
悋惺惶惡悋惠悋悽惘悴悋惠愕悋惠悋惠惺
悋惠悋愕忰惆惓惠愆愀惺惶惡悋忰惆愀愕
悋惠.愕惺惶惡悋悽惘悴悋惠悋悵愕惡
悋悋愕悸惡惺惶惡悋悋愕惡悋悵悖悽悵悋
(winner-takes-all)
Dr gafar zen alabdeen salh
(2011) 2
惴惘惠悋惘悸悋悖愕悋愕悸惠惺悋惠悋愕惡惆悋悸
悋愕惡惺悋惠悋惘悋惺愆惘悋悋惆(悴惘愕惡惘悴
19751973愆悋1975)悒悋悖悋惠惺
悋惠悋愕愆惆悋惠悋悋惡惘悋忰惠悋悸
悋惓悋悋惠悋惘悋惺愆惘悋悋惆惺惆悋惆
惠teuvo kohnen愀惡悸悽悋惶悸
悋愆惡悋惠悋惺惶惡悸惠愕忰悋惠愕悸悋惠惴
悋悵悋惠self-organizing feature map(SOM)
1989.惡惠悵悋悽惘悋悧愀惺悋惠惺
悋惠悋愕.
Dr gafar zen alabdeen salh
(2011) 3
惠惺
KOHNEN learning
self-organizing map
(SOM)
Dr gafar zen alabdeen salh
(2011) 4
悒惺惆悋惆:惆.惶悋忰 悋惺悋惡惆 慍 悴惺惘
悋惺悋惠 惠悋悸 悋忰悋愕惡 惺 悸 悋 悴悋惺悸
The Applications of Kohonen
Network
 1. Visualization aids
 2. Dimension reduction of large, high
 dimensional data sets
 3. Classification purposes
 4. Speech recognition
 5. Surface reconstruction in Computer
 Graphics
Dr gafar zen alabdeen salh
(2011) 5
 悋悵悋惠 悋惠惴 忰 惠惓  
惺惶悋愃悸悋惺惆悸惠愆悋忰
悋愀惡愃惘悋principle of topographic map
formation(1990).惠悵惘悵悋悋惺惆悸悖
悋惺悋悋spatial location惺惶惡
悋悽惘悴悋惠悋愀惡愃惘悋悋惴惘愕悸惺悸愀
悋惆悽悋惠.悋悋惠惘忰悖惷悋悵悴惠忰
悋愕悸悋惡悋愆悋惠悋.愕惠悽惶悵悋
悋悵悴悋愕悋惠悋惘悧愕悸忰悋惠悋惠惴悋悵悋惠
悋悽惠惓悋惡惺惆悵悋忰悋愕惡
惡愕悸.
Dr gafar zen alabdeen salh
(2011) 6
Feature-mapping Kohonen model
Dr gafar zen alabdeen salh
(2011) 7
Input layer
Kohonen layer
(a)
Input layer
Kohonen layer
1 1
(b)
00
Dr gafar zen alabdeen salh
(2011) 8
Dr gafar zen alabdeen salh
(2011) 9
Dr gafar zen alabdeen salh
(2011) 10
惘悵悴忰悋愀惡愃惘悋悋.惷惺惘悋
惓悋惡惠悋悖悋愀悋惆悽悋惠愀惡悸悋惆悽悋惠悽惘悴悋惠
悖惡惺悋惆悋惺悖愀惡悸.悋愆悋愕悋惡惠惠
愀惡悸愆惡悸LATTICE惡惺惆悸4
4惺惶惡,惺惶惡悋惓悋悋惆悽悋惠.
惡悋惺惶惡悋悋悧慍惡悋悋悖愕惆悴惘悋惡悋
悋惘悋惆悋悴惘悋悋悋悧慍惺惡悋惘悸惺惺惶惡悋惠惘惡
惘惡悋愀惡惺悋悋悋悧慍
Dr gafar zen alabdeen salh
(2011)
11
惘惡 惆 悋悋愀惡惺 悋惘惡
惠忰惆惆惆悋惘惡悋愀惡惺惡悋愕愀悸惶
悋愆惡悸悖愆悴惘悋悋悋悧慍
惺惶惡悋惠悋惺悋忰惆悖惺悖忰惠
惓悋惓悸悋惺惺悴悋惡.惓悋悵惓
愆悋愕悋惡悴惘悋悋悋悧慍悋忰悴悋忰惆.
惡惶悸惺悋悸惡惆悖悋惠惆惘惡愆惡悸
惡悴惘悋悋悋悧慍悋忰悴悋惡惘.惡惺惆悵
愕惠惘悋惠惆惘惡惺悋悽悋惷忰悴悋悴惘悋
惠惆惘悴悋
Dr gafar zen alabdeen salh
(2011) 12
惠惠愆惡悸愀惡悸悋忰惆悸惺惶惡悋惠
悋忰愕悋惡悋惠悋惺悋悽惠悋悋悋惘惠惡悋愀悋惠
悋悋惘惠惡悋愀悋惠悖悋(forward connection)
悋惺惶惡悋惠愀惡悸悋惆悽悋惠悒悋惺惶惡悋惠愀惡悸
悋悽惘悴悋惠悵悴悋惡悸(lateral connection)惡
悋惺惶惡悋惠愀惡悸悋悽惘悴悋惠
惠愕惠悽惆悋悋惘惠惡悋愀悋惠悋悴悋惡悸悒惠悋悴惠悋愕惡
悋惺惶惡悋惠.惶惡忰悋惺惶惡悋悵悋惡惘愕惠惠愆愀
惺惡惘悋惺惶惡悋惠愀惡悸悋悽惘悴悋惠悋悋悧慍(惺惶惡
悋悋悧慍悖悽悵悋winner-taked-all-neuron)
悵悋悋惺惶惡悋忰惆悋悵惠悴悒愆悋惘悸悽惘悴悋惠.愃
愆悋愀悋惺惶惡悋惠悋悋悽惘悋悋愕悸
Dr gafar zen alabdeen salh
(2011) 13
惺惆惠惆愀惆悽悋惠悋愆惡悸愕惠惡
惺惶惡愀惡悸愕悽悸悋悸愀
悋惆悽悋惠惠惠惺惆惡悋愕愀悸愕悋惘悋悽悋悖慍悋
悋惘惠惡悋愀悋惠悋愀悋悋愆惠惡悋惡愀惡悋惠悋惆悽悋惠
愀惡悸惠惠悴悋惘惠惡悋愀悋惠悋惠愃悵悸悋惘惠悴惺悸
悋悴悋悸悋惠悖惓惘惓惘悸悖悋惺悸悋惺惠悋惆悋惺悋愕悋悸
悋惺惶惡悋悋悧慍.惠忰悵悋惺愀惘悋愕惠悽惆悋
惆悋悸悋惡惺悸悋愕悸
Dr gafar zen alabdeen salh
(2011) 14
Mexican hat function
悋愕悸 悋惡惺悸 惆悋悸
惠惓悵悋惆悋悸悋惺悋悸惡悋愕悋悸惡惺惶惡
悋悋悧慍悖悽悵悋悸悋悋惘惠惡悋愀悋惠愀惡悸.
愀惡悋悵悋惆悋悸悖惘惡悋悴惘悋(愀悸悒惓悋惘悸
悴悋惡悸惶惘悸悖惆)惠悖惓惘悒惓悋惘悸悴悋惘
悋惡惺惆(愆惡悋惴悋悋惺an inhibitory)惠悋惓惘悋惺
惺惠惆
悴悋惘悋惡惺惆悴惆悋(悋愀悸悋忰愀悸惡愆惡悸
悋惴悋悋惺)惠悖惓惘悒惓悋惘悸惷惺悋悵悋惺悋惆悸.
悋惴惘悋愆悋惠悋
Dr gafar zen alabdeen salh
(2011) 15
悋愕悸 悋惡惺悸 惆悋悸
The Mexican hat function of lateral connection
Dr gafar zen alabdeen salh
(2011) 16
Connection
strength
Distance
Excitatory
effect
Inhibitory
effect
Inhibitory
effect
0
1
愆惡悸惠惺悋惺惶惡惺愀惘惠惘忰
悖慍悋悋惘惠惡悋愀悋惠愃惘愆愀悸悒悋惘惠惡悋愀悋惠愆愀悸.
愕忰惺惶惡悋悋悧慍悴惘悋愀惠惺.悒悵悋
愕惠悴惡悋惺惶惡愀惆悽悋惠惺惺惆悵悋
悖忰惆惓悋惠惺悵悋悋惺惶惡悋忰惆惆.
惠惷惺悒愆悋惘悸悋悽惘悴悋惠yj惺惶惡悋悋悧慍悖悽悵悋j
惠愕悋1惠惷惺悒愆悋惘悸悽惘悴悋惠悋惺惶惡悋惠悋悋悽惘
(悋惠悽愕惘惠悋悋愕悸)惠愕悋0.
惠惺惘悋惺惆悸悋惠惺悋惠悋愕悋愀standard
competitive learning rule(悋1994)悋惠愃惘
悋慍悋悵愀惡惺慍悋愀悋悋愆惠惡悋...悋
:
Dr gafar zen alabdeen salh
(2011) 17
Dr gafar zen alabdeen salh
(2011) 18
忰惓xi悒愆悋惘悸悋惆悽悋惠留惺悸惺惆悋惠惺惠惺惡0悒
1
惺悋惠悖惓惘悋愆悋惠惺悋惠悋愕悋惠悋惠悴慍悋悋愆惠惡悋
wj惺惶惡悋悋悧慍j惠悴悋愀悋惆悽悋惠x.惺悋惘
悋悋惠悋悋悧悋悖愕悋悸悋惆悸Euclidean distance惡
悋惠悴


 
緒
ncompetitiothelosesneuronif,0
ncompetitiothewinsneuronif),(
j
jwx
w
iji
ij
悋悋惆悸 悋愕悋悸 悋
惠惺惘悋愕悋悸悋悋惆悸惡慍悴惠悴悋惠xwi
(1n)惡悋惺悋悸
忰惓xiwij悋悋惺惶惘悋悋惠悴xwj
惺悋惠悋
Dr gafar zen alabdeen salh
(2011) 19
2
1
1
2
)( 





緒 ワ
n
i
iji wxwjxd
惠忰惆惆悋惠愆悋惡惡悋惠悴xwj惡悖惠惡悋惆
愕悋悸悋悋惆悸d惠惓悋愕悋悸悋悋惆悸惡
悋惠悴xwj愆悋惠悋惡愀悋悽愀悋悵
惶惆悸悋惠悴惠惷忰悋愆惠悋悋悋
悋惶愃惘惠悋愕悋悸悋悋惆悸悋悵悋惆悋惠愆悋惡惡
悋惠悴xwj
Dr gafar zen alabdeen salh
(2011) 20
悋悋悧慍 悋惺惶惡  
惠惺惘惺悋惺惶惡悋悋悧慍jx悵悋悵
悖惷悋惠悋惺惠悴悋惆悽悋惠x悋惠愀惡
悋愆惘愀悋惠悋(悋1994):
忰惓m惺惆惆悋惺惶惡悋惠愀惡悸

Dr gafar zen alabdeen salh
(2011) 21
mjwjxjx ,...,2,1,min 緒
惓悋:
惡惺惆  悋惆悽悋惠 惠悴 惠惆 惠 悋 悋惘惷x
惺惶惡悋惠 惓悋惓悸 惡悋  愆惡悸:
 悋 悋悋惡惠惆悋悧悸 悋慍 惠悴悋惠 惠惺愀:
Dr gafar zen alabdeen salh
(2011) 22







12.0
52.0
x






緒





緒






21.0
43.0
,
70.0
42.0
,
81.0
27.0
321 www
悋悋悧慍 悋惺惶惡 悴惆jx(悋惠悋 悖惷 惶悋忰惡)
悋惆悸 愕悋悸 悋 惺悋惘 惡悋愕惠悽惆悋:
悵悋悋惺惶惡3悋悋悧慍惠悴惆惆惠悴慍
w3愀惡悋悋惺惆悸悋惠惺悋惠悋愕悋惶悸惡悋惺悋惆悸
悋惠悋悸.惡悋惠惘悋惷悖惺悸惺惆悋惠惺留=0.1
悒悋忰惶惺悋:
Dr gafar zen alabdeen salh
(2011) 23
13.0)21.012.0()43.052.0()()(
59.0)70.012.0()42.052.0()()(
73.0)81.012.0()27.052.0()()(
222
232
2
1313
222
222
2
1212
222
212
2
1111
緒緒
緒緒
緒緒
wxwxd
wxwxd
wxwxd
悋 惺 忰惶:
悋悴惆惆 悋慍 惠悴 惠忰惆惆w3悋惠惘悋惘 惺惆p+1 悋:
Dr gafar zen alabdeen salh
(2011) 24
01.0)21.012.0(1.0)(
01.0)43.052.0(1.0)(
23223
13113
緒緒緒
緒緒緒
wxw
wxw








緒












緒緒
20.0
44.0
01.0
01.0
21.0
43.0
)()()1( 333 pwpwpw
悋惠悋愕 悋惠惺 悽悋惘慍悸
悋悋惡惠惆悋悧悸 悋惷惺
悋惠愆悋惡惠 悋惠愆愀:
悋悋悧慍 悋惺惶惡 悒悴悋惆
悋惠惺:
悋慍悋悖 惠悴惆惆
悋惘惘悋惠
Dr gafar zen alabdeen salh
(2011) 25
Dr gafar zen alabdeen salh
(2011) 26
悋惠悋愕 悋惠惺 悽悋惘慍悸
悋悋惡惠惆悋悧悸 悋惷惺
悖慍悋悋愀悋悋愆惠惡悋惡惺愆悋悧悸惡0悒1
悸悴惡惶愃惘悸惺悸惺惆悋惠惺留

Dr gafar zen alabdeen salh
(2011) 27
悋惠愆悋惡惠 悋惠愆愀:悋悋悧慍 悋惺惶惡悒悴悋惆
惠愆愀愆惡悸惺愀惘惠愀惡惠悴悋惆悽悋惠X悒悴悋惆悋惺惶惡JX
悋悋悧慍悖悽悵悋(WINNER TAKE ALL)悋悵悖惷悋惠悋惺惆
悋惠惘悋惘P惡悋愕惠悽惆悋惺悋惘悋愕悋悸悋惆悸
忰惓n 悋惆悽悋惠 惺惶惡悋惠 惺惆惆m悋悽惘悴悋惠 愀惡悸  悋惺惶惡悋惠 惺惆惆悋
愀惡悸
Dr gafar zen alabdeen salh
(2011) 28
mjwjxjx ,...,2,1,min 緒 
2
1
1
2
)( 





緒 ワ
n
i
iji wxwjxd
Dr gafar zen alabdeen salh
(2011) 29
悋惠惺:惠悴惆惆悋慍悋悋
Dr gafar zen alabdeen salh
(2011) 30
 )()),((
)(,0)(
pjpwX
pjij
jiji
j
pw

緒

)()()1( pwpwpw ijijij 緒
悋惠悋愕 悋惠惺 悋惺惆悸 惡悋愕愀悸 悋慍 惠惶忰忰 惠忰惆惆
忰惓留 悋惠惺 惺惆 惺悸j(p)忰 悋惠惘慍悸 悋悴惘悸 惆悋悸
悋惺惶惡jx悋惠惘悋惘 惺惆 悋悋悧慍p
惓悋惡惠 悋惘惠悋惺 悋悴惘悸 惆悋悸  惺悋惆悸.悋惺惶惡  悖 惠愆悋惠
悋忰惆 惠  惠愆愀 悋愀惡愃惘悋悸 悋悴惘悸 惆悋悽 悋悋惺悸.悋惺惆
悋悖愆悋
 悋悋惺惶惡 悽惘悴悋惠 惠惺惘 :
Dr gafar zen alabdeen salh
(2011) 31
 )(,1
)(,0
Pj
Pjj
j
j
y
Dr gafar zen alabdeen salh
(2011) 32
Dr gafar zen alabdeen salh
(2011) 33
Dr gafar zen alabdeen salh
(2011) 34
Dr gafar zen alabdeen salh
(2011) 35
Dr gafar zen alabdeen salh
(2011) 36
Dr gafar zen alabdeen salh
(2011) 37
Dr gafar zen alabdeen salh
(2011) 38
悋惠惘悋惘
悖惷1悒悋惠惘悋惘p悋惘悴惺悒悋悽愀悸悋惓悋悸
悋愕惠惘忰惠惠忰惺悋惘悋愕悋悸悋惆悸悖悋
悖悋惠忰惆惓惠愃惘悋惠忰惴悸惠忰悋愕悸
Dr gafar zen alabdeen salh
(2011) 39
Initial random weights
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1-1
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
W(2,j)
W(1,j)
Network after 100 iterations
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8-1
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1
W(2,j)
W(1,j)
Network after 1000 iterations
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8-1
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1
W(2,j)
W(1,j)
Network after 10,000 iterations
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8-1
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1
W(2,j)
W(1,j)
Dr gafar zen alabdeen salh
(2011) 44
The end

More Related Content

Som kohnen learning

Editor's Notes