cppn2

6
How do CPPNs work?

Upload: warunikaranaweera

Post on 01-Feb-2016

212 views

Category:

Documents


0 download

DESCRIPTION

Computational pattern producing networks.

TRANSCRIPT

Page 1: CPPN2

How do CPPNs work?

Page 2: CPPN2

Steps

• Step 1: Randomly generate a genome

• Step 2: Pass the genome through a set of predefined functions to get the phenotype

• Step 3: Calculate fitness

• Repeat step 2 and step 3 if newFitness > lastFitness

Page 3: CPPN2

Step 1: Random generation of genomes

Example: 256*256 pixels

Assign some random number to each pixel

Page 4: CPPN2

Step 2: Express the phenotype

• Predefined functions (hardcoded):

– E.g.: If the value at a pixel is 2 new pixel value is Sin(2)

• Phenotype => new color of that pixel

GenotypeCPPN

Phenotype

Page 5: CPPN2

Step 3: Calculate the fitness

• Fitness is calculated here to test whether the actual image is closer to the phenotype

• Fitness = Expressed color – Actual color (at one pixel position)

• Repeat step 2 and step 3 if

– currentFitness > lastFitness

Page 6: CPPN2

Conclusion

• Functions in the CPPN are hardcoded (fixed for a set of images)

• From initial random pixel values , a CPPN can produce the actual image

• By mutating (i.e. weight change) or crossing of two CPPNs, we can build a new CPPN (I am yet to verify this)