Professional Documents
Culture Documents
Evolving Neural Networks Through Augmenting Topologies
Evolving Neural Networks Through Augmenting Topologies
Evolving Neural Networks Through Augmenting Topologies
through Augmenting
Topologies
Daniele Loiacono
24/04/2008
Daniele Loiacono
Neuroevolution
Daniele Loiacono
Outline
Motivations
Neuroevolution
NEAT
Applications
Daniele Loiacono
Motivations to neuroevolution
Daniele Loiacono
Design of network topology
Daniele Loiacono
Problem Dimensionality
Daniele Loiacono
High Dimensional Space is Hard to Search
A rough idea:
• 3 dimensional – easy
• 100 dimensional – need a good optimization method
• 10,000 dimensional – very hard
• 1,000,000 dimensional – very very hard
• 100,000,000,000,000 dim. – forget it
Unfortunately, most interesting problems have
solutions in high dimensional space
How to get there?
Daniele Loiacono
Outline
Motivations
Neuroevolution
NEAT
Applications
Daniele Loiacono
Evolutionary Computation
Several approaches:
Daniele Loiacono
The population
Selection
probability is
usually
proportional
to fitness
Exploit goodness
of
parents
Explore
unknown
Daniele Loiacono
Genotype and phenotype
Daniele Loiacono
Genotype and phenotype
Genotype
Phenotype
Daniele Loiacono
Genotype and phenotype
Genotype
Phenotype
f ( x) = 3x + 7 x − 10
2
Daniele Loiacono
Genotype
<x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12>
Phenotype
Daniele Loiacono
Fitness
Daniele Loiacono
Selection
Daniele Loiacono
Mutation and Recombination
Mutation
• changes the genotype randomly
• source of diversity and discovery
• a kinf of “local search”
Recombination
• combines different genotypes
• should be aware of structures of solution !
Daniele Loiacono
Diversity
Daniele Loiacono
Speciation
Daniele Loiacono
Building Blocks
x x
Competent GAs
Daniele Loiacono
Building Blocks
Daniele Loiacono
Outline
Motivations
Neuroevolution
NEAT
Experimental results
Daniele Loiacono
Neuroevolution (NE)
Daniele Loiacono
TWEANN issues
Encoding
• direct
• indirect
Mating (Crossover)
• competing conventions
• free topology and the Holy Grail
Protecting Innovations
Initialization and topology minimization
Examples of TWEANN systems
Daniele Loiacono
TWEANN Encoding
Daniele Loiacono
Mating in TWEANN
n1 ļ n’1
n1 n2 n3 n’1 n’2 n’3
; n2 ļ n’2
n3 ļ n’3
Daniele Loiacono
Mating in TWEANN
A B CA C B C A
…
Competing Conventions Problem
Daniele Loiacono
Mating in TWEANN
Daniele Loiacono
Protecting innovations
Daniele Loiacono
Initialization and topology minimization
Daniele Loiacono
Structured Genetic Algorithm (sGA) Dasgupta and
McGregor (1992)
Daniele Loiacono
Breeder Genetic Programming (Zhang and Muhlenbein)
Daniele Loiacono
Parallel Distributed Genetic Programming (PDGP)
Pujol and Poli (1997)
Dual encoding:
• linear genome for mutating weights
• Graph representation for a subgraph swapping
crossover
Does swapping crossover guarantee a good
mating?
Daniele Loiacono
GeNeralized Acquisition of Recurrent Links (GNARL)
Angeline, Saunders, and Pollack (1993)
Graph Encoding
Avoid mating because considered disruptive
Daniele Loiacono
Cellular Encoding
Gruau (1993, 1996)
Daniele Loiacono
Enforced SubPopulations (ESP)
Gomez and Miikkulainen (1997,1999)
Fixed topology
When it fails starts from scratch with a new
random topology
Faster than Cellular Encoding!
Daniele Loiacono
Outline
Motivations
Neuroevolution
NEAT
Applications
Daniele Loiacono
NEAT
Daniele Loiacono
Genetic Encoding in NEAT
Daniele Loiacono
Topological Innovation
Daniele Loiacono
Link Weights Mutation
Daniele Loiacono
Topology Matching Problem
Daniele Loiacono
Historical Markings
Daniele Loiacono
Speciation in NEAT
Daniele Loiacono
Network similarity in NEAT
Daniele Loiacono
Fitness Sharing in NEAT
Daniele Loiacono
Complexification
Daniele Loiacono
Outline
Motivations
Neuroevolution
NEAT
Applications
Daniele Loiacono
NEAT Applications
Daniele Loiacono
Conclusions
Daniele Loiacono
XCSF with Neural Prediction
6WDWH$FWLRQ *$RQ11
6SDFH 11
11
11
11
11
11
11
11
*$
Daniele Loiacono
XCSF with Neural Prediction
6WDWH$FWLRQ *$RQ11
6SDFH 11
11
11
11
11
11
11
11
*$
Daniele Loiacono
XCSF with Neural Prediction
6WDWH$FWLRQ *$RQ11
6SDFH 11
11
11
11
11
11
11
*$
Daniele Loiacono
References
Daniele Loiacono