Documentos de Académico
Documentos de Profesional
Documentos de Cultura
COMPUTER VISION
ndice
Aprendizaje no supervisado: Mapas Auto-organizados SOM - Inspiracin y objetivos - Aprendizaje - Ejemplo y aplicaciones
1
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
Cada neurona debe aprender un patrn Neuronas cercanas deben aprender patrones similares
Rn
R2 , R1
2
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
La relacin de vecindad de las neuronas est predefinida Los pesos de cada neurona determinan su posicin en el espacio de entrada
Ejecucin:
qu neuronas se activan ante cada entrada?
Aprendizaje:
qu neuronas aprenden la nueva muestra? cmo la aprenden?
3
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
ndice
Aprendizaje no supervisado: Mapas Auto-organizados SOM - Inspiracin y objetivos - Aprendizaje - Ejemplo y aplicaciones
4
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
ds(wi-wg)
k
5
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
Solucin neuronal:
P. Campoy
CVG-UPM
COMPUTER VISION
Ejemplos de SOM de R2 en R1
caracterstica 2
caracterstica 1 caracterstica 2
caracterstica 1
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
ndice
Aprendizaje no supervisado: Mapas Auto-organizados SOM - Inspiracin y objetivos - Aprendizaje - Ejemplo y aplicaciones
8
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
Ejemplos SOM
9
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
Ejemplos SOM
10
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
v=0.25
v=0.5
v=0.75
11
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
SOM result
different order
P. Campoy
12
CVG-UPM
COMPUTER VISION
Comandos Matlab
som1=newsom(minmax(psom),[10 1]); som1=train(som1,psom) ynt=sim(som1,tsom); plotsom(som1.iw{1,1},som1.layers{1}.distances)
13
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
p.valor
t.valor
14
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
C1 C1 C2
P. Campoy
C2 11 182 C1 C2
C1 101 6
C2 6 187
15
100 7
CVG-UPM
COMPUTER VISION
16
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
17
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
18
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
Entrada 5 dimensiones: % de H2 , CH4 C2H2 C2H4 C2H6 Salida bidimensional con U-matriz Semntica manual supervisad
19
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
P. Campoy
20
CVG-UPM
COMPUTER VISION
21
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
H=0.4375
22
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
I(E)
Ejemplo:
P(E)
Dado con 8 caras equiprobables con valores: 1,1,1,1,2,2,3,4 suceso P(ai) I(ai)=-log2P(ai) cdigo 1 0,5 1 0 2 0,25 2 10 3 0,125 3 110 4 0,125 3 111
P. Campoy
CVG-UPM
COMPUTER VISION
Limitaciones:
- muestras cercanas activan neuronas distantes - neuronas vecinas en pueden ser activadas por muestras distantes
24
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
Ejemplos SOM
CVG-UPM
COMPUTER VISION
dim. intr. 2
dim. intr. 1
26
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
dim. intr. ?
27
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
12x12
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0 0.1 0.2 0.3 0.4 0.5 0.6 Red 12x12 0.7 0.8 0.9 1
0.9
1x144
5x29
0 0.1 0.2 0.3 0.4 0.5 0.6 Red 5x29 0.7 0.8 0.9 1
28
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
SOMOS
29
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
SOMOS
30
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
SOMOS
31
P. Campoy Neural Networks and Pattern Recognition
CVG-UPM
COMPUTER VISION
SOMOS
32
P. Campoy Neural Networks and Pattern Recognition