Please enable JavaScript.
Coggle requires JavaScript to display documents.
Evolutionary Algorithm Variants (Evolutionary Programming (Parent …
Evolutionary Algorithm Variants
Simple
Genetic Algorithm
Shortcomings
Mutation & Crossover only applicable
for bit-string & integer representations
Selection mechanism sensitive
for converging populations with
close fitness values
Representation too restrictive
Generational population model can
be improved with explicit survivor selection
Using crossover
AND
mutation sequentially
Evolution Strategies
Mutation
σ is varied on the fly by the
1/5
success rule
Rule resets σ
after every k iterations by:
σ = σ / c if p>1/5
σ = σ * c if p<1/5
σ = σ if p = 1/5
p is the % of successful mutations,
0.8 <= c <= 1.0
z values drawn from
normal distribution N(u, σ)
u = mean
σ = variation/ mutation step size
Chromosome consists
3 vectors
Object variable, x
Mutation step size
Rotation angle, α
Parent
Selection
Uniform
random selection
Selection is
unbiased
Crossover
Local Discrete
z = x OR y
Two fixed parents
Global Intermediary
z = x OR y
Two parents selected for each i
Global Intermediary
z = (x + y) / 2
Two parent selected for each i
i represent gene index
Local Intermediary
z = (x + y) / 2
Two fixed parents
Survivor Selection
(μ,λ) / (μ+λ)
Evolutionary Programming
Aim at
artificial intelligence
Capability to
Predict
Prediction by finite state machine
Chromosome consists
2 vectors
Object variable, x
Mutation step size
Mutation
Gaussian Perturbation
No Crossover
Parent
Selection
Each individual creates
one child
by mutation
Deterministic
&
unbiased
by fitness
Survivor Selection
(μ+μ)
Genetic Programming
Using crossover
OR
mutation with probabilistic
Attributed Features
Needs
huge
populations
Slow
Competes with
neural nets
and alike
Tree structure
Crossover
Exchange of subtrees
Mutation
Random change in tree
Survivor Selection
Generational replacement
Parent Selection
FPS