02 GeneticAlgorithm

Genetic Algorithm
Profesor: Alexandre Bergel

Auxiliar: Juan Pablo Silva
abergel@dcc.uchile.cl
jpsilva@dcc.uchile.cl
2
Premise of Genetic Algorithm
Natural selection, as in biological systems, is very

successful at optimizing individual and populations
If we can mimic natural selection, then we will be able
to optimize a function over a set of individual elements
Any element can be considered as an individual who
is fighting for its survival within a large population
3
Premise of Genetic Algorithm
Natural selection is pioneered by Charles Darwin

(1809-1882), biologist who worked on natural
evolution
Controversial and influential book by Darwin
“On the Origin of Species by Means of Natural
Selection, or the Preservation of Favoured Races in the
Struggle for Life”, 1859
“One general low, leading to the advancement of all
organic beings, namely, vary, let the strongest live
and the weakest die”
4
— Charles Darwin
“Guess the 3-letter word I have in mind”
Secret word: cat

5
“Guess the 3-letter word I have in mind”
“For each try, I tell you the number of correct letters”
Secret word: cat

6
“cow”
1
“poz”
0
“gat”
Secret word: cat

7
gat gaz
cow gow
Using genetic
operators,
the words are
combined and
some letters are
randomly modified
8
gat gaz gat
cow gow cat
9
gat gaz gat
cow gow cat
10
1st generation 2nd generation
gat gaz gat

cow gow cat
11
Flow chart of an evolution algorithm
Selection
Population Parents
Reproduction
Replacement
Offspring
12
Flow chart of an evolution algorithm
Selection
Population Parents
Reproduction
Replacement
Many EAs are around: Offspring
Ant colony optimization,
Artificial immune system,
Cultural algorithms,
Genetic Algorithm,
Genetic Programming,
Grammatical evolution,
… 13
Flow chart of a genetic algorithm
Initialize
population
Evaluate fitness Reproduction
Solution No
Selection
found?
Yes
End
14
Flow chart of a genetic
We create algorithm
a
set of random 3-letter
Initialize words
population
Solution No
Selection
found?
Yes
End
15
Initialize Compute the
population
score of each word
Solution No
Selection
found?
Yes
End
16
Initialize
population

Do we have a word with
score = 3?
Solution No
Selection
found?
Yes
End
17
Initialize
population

We take the words that
are close to 3
Solution No
Selection
found?
Yes
End
18
We combine and
Initialize mutate words to form a new
population set of words
Solution No
Selection
found?
Yes
End
19
Genetic Algorithm in a Nutshell
Evolutionary computation technique that automatically

solves problems without specifying the form or structure
of the solution in advance
Generally speaking, genetic algorithms are simulations
of evolution, using biological genetic operations
May be combined with other optimization techniques
20
Genetic Algorithm in a Nutshell
The idea first appears in 1967, in J. D. Bagley’s thesis

“The Behavior of Adaptive Systems Which Employ
Genetic and Correlative Algorithms”
Since then, this field has seen a tremendous
development
GA considered as an optimization method
ie. finding x such as f(x) = y is maximal, x ∈ X , a
multidimensional space
21
Terminology: Individual
An individual represents an element in a population of

chromosomes.
Each individual has a chromosome, composed of
genes
Gene = index in a chromosome

Allele = value of a gene
Individual
22
Terminology: Population
A population is a set of individuals

Generally, the population size is fixed over time.
Individuals are replaced at each generation, but the
number of individuals remains constants.
All individuals of the population have the same size
23
Terminology: Fitness function
The fitness function evaluates how fit an individual is
f( )=y y∈ℝ
The whole idea of genetic algorithm is to search for

the individual that maximizes the fitness function
24
Example: optimizing a server
Consider a server running in Java

The Java virtual machine, which has over 200 options
What are the options that consume the least amount
of memory to answer HTTP requests
f(x) = y
x is a set of options, e.g., [ -Xgc:parallel, -ms32m,
-mx200m]
y is the amount of memory (in bytes)
f(x) is computed by launching the server using the

options and sending 1000 requests
25
Example: optimizing a server
Consider a server running in Java

The Java virtual machine, which has over 200 options
We look for the optimal x
What are that
the options thatf(x)
minimize consume the least amount
of memory to answer HTTP requests
f(x) = y
x is a set of options, e.g., [ -Xgc:parallel, -ms32m,
-mx200m]
y is the amount of memory (in bytes)
f(x) is computed by launching the server using the

options and sending 1000 requests
26
Example: space antenna
Evolved antenna, produced by
NASA in 2006.
Used in 3 satellites that take
measurement Earth
magnetosphere.
Satellites used this antenna to
communicate with the ground
27
Example: space antenna
f(x) = y
List of commands: voltage and impedance

forward(length, radius)
for radio waves
rotate-x(angle)
rotate-y(angle)
rotate-z(angle)
https://ti.arc.nasa.gov/m/pub-archive/1244h/1244%20(Hornby).pdf
28
Example: software testing
f(x) = y
Actions on the user interface Number of tested

(e.g., pressing a button, entering a functionalities
value in the text field, clicking on a
menu)
Randomly testing an application has many

applications:
Finding functional bugs (i.e., which actions crashes my application)
Finding dead code (i.e., which part of my application is not used)
29
Flow - chart of a genetic algorithm
Initialize
population
Solution No
Selection
found?
Yes
End
30
Selection phase
Several selections algorithms have been proposed

The two most popular are
Roulette associates a probability of selection to each individual
Tournament between a set of randomly picked individuals
31
Roulette selection
i3
i1 (17%)
f(i1)=200 (35%)
f(i2)=150 i2
f(i3)=100 (26%)
f(i4)=120 i4
(21%)
Selection point
32
Roulette selection
If fi is the fitness of an individual i, then its probability

of being selected is fi
pi =
∑j fj
This can be simulated by the following (naive) algorithm:

1 - Calculate the sum of all fitnesses in population (sum S).
2 - Generate a random number r in the interval [0; S].
3 - Go through the population and sum fitnesses. When the
sum s is greater than r, stop and return the individual where
you are.
33
Tournament
The tournament selection algorithm:

1 - Choose few individuals (e.g., 5) at random from
the population (a tournament)
2 - The individual with the best fitness (the winner) is
selected for crossover
The main advantage is to be highly parallelizable

across multiple CPU Cores
34
Initialize
population
Solution No
Selection
found?
Yes
End
35
Darwinian Natural Selection
In order to have a natural selection, we need to have:

Heredity: a child receives properties of its parents. In particular, if
the parents are robust and can live long enough, then the child
should too
Variation: some variation may be introduced in children. Children

should not be identical copy of their parents
Selection: some members of a population must have the

opportunity to be parents and have oﬀsprings in order to pass their
genetic information. Typically referred to as “survival of the fittest”
https://en.wikipedia.org/wiki/Charles_Darwin#/media/File:Charles_Darwin_seated_crop.jpg
36
Genetic operators: Crossover
Operations that takes two individuals and produces a

new one, result of a combination
37

Gene randomly chosen

38

39

New individual
40
Genetic operators: Variant of
Crossover

Two new individuals
41
Genetic operators: Mutation

42

Gene randomly chosen
43

New random allele
44
Initialize
population
Solution No
Selection
found?
Yes
End
45
Fitness function
GA searches for a solution, without specifying an

analytic way of doing so
This is very appealing since we just need to say how
to evaluate a possible solution instead of saying how
to get it
However, writing a fitness function is complex
It happens that often some constraints have to be
specified:
E.g., a robot needs to find the exit and find the shortest path
E.g., numbers cannot be repeated when solving the knapsack

problem
46
Fitness function
Constraints may be expressed using different ways:

Multi-objective fitness function
Penalty for violation in the fitness function
Implicit via the encoding of the problem
Implement a repair capability in case of an infeasible / unsound

individual
47
Minimization vs Maximization
For some problems, getting closer to the solution

means decreasing a function. E.g.,
Distance between a robot to the exit
Error function
In such a case, the fitness function needs to be

expressed as:
- objective
N - objective
1 / (1 + objective)
48
Encoding / decoding
Genotype Phenotype
TAAGTG
(biology)
‘C’, ‘A’, ’T' ‘CAT’
‘north’, ‘south’, ‘south', …
49
Encoding / decoding
Genotype Phenotype
2
1, 3, 2, 4, 5 1
5
3
forward(2, 30), rotate-y(20), …
50
Configuring the algorithm
Mutation rate: % to change a gene when creating a

child
Population size: number of individual to consider each
time
Number of genes: how many genes contains each
individual
Fitness function: Function that tells how good / far an
individual is from the (ideal) solution
51
Result of the algorithm
Once the algorithm is run, we need to know how

good we did
At least two metrics are important to have:
Number of generations until the solution is found
Total time until the solution is found
52
Some benefits of Genetic Algorithm
GA provides a compelling way to not be trapped in

local optima
GA allows optimization of systems in which variables
may be discrete or categorical, and not only
continuous
e.g., direction of a robot or characterization of an antenna segment
53
What have we seen?
Complete description of a basic infrastructure to run

genetic algorithm
A few non-trivial examples of genetic algorithm
54
Practical session
The colab notebook is available on:

http://bit.ly/TallerAI-Session2
55

02 GeneticAlgorithm

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

02 GeneticAlgorithm

Uploaded by

Copyright:

Available Formats

Genetic Algorithm

Profesor: Alexandre Bergel

Natural selection, as in biological systems, is very

Natural selection is pioneered by Charles Darwin

Secret word: cat

Secret word: cat

Secret word: cat

gat gaz gat

Evaluate fitness Reproduction

Evaluate fitness Reproduction

Evaluate fitness Reproduction

Evaluate fitness Reproduction

Evaluate fitness Reproduction

Evaluate fitness Reproduction

Evolutionary computation technique that automatically

The idea first appears in 1967, in J. D. Bagley’s thesis

An individual represents an element in a population of

Gene = index in a chromosome

A population is a set of individuals

The fitness function evaluates how fit an individual is

The whole idea of genetic algorithm is to search for

Consider a server running in Java

y is the amount of memory (in bytes)

f(x) is computed by launching the server using the

Consider a server running in Java

y is the amount of memory (in bytes)

f(x) is computed by launching the server using the

List of commands: voltage and impedance

Actions on the user interface Number of tested

Randomly testing an application has many

Finding dead code (i.e., which part of my application is not used)

Evaluate fitness Reproduction

Several selections algorithms have been proposed

Tournament between a set of randomly picked individuals

If fi is the fitness of an individual i, then its probability

This can be simulated by the following (naive) algorithm:

The tournament selection algorithm:

The main advantage is to be highly parallelizable

Evaluate fitness Reproduction

In order to have a natural selection, we need to have:

Variation: some variation may be introduced in children. Children

Selection: some members of a population must have the

Operations that takes two individuals and produces a

Operations that takes two individuals and produces a

Gene randomly chosen

Operations that takes two individuals and produces a

Operations that takes two individuals and produces a

Operations that takes two individuals and produces a

Two new individuals

Operations that takes two individuals and produces a

Operations that takes two individuals and produces a

Operations that takes two individuals and produces a

New random allele

Evaluate fitness Reproduction

GA searches for a solution, without specifying an

E.g., numbers cannot be repeated when solving the knapsack

Constraints may be expressed using different ways:

Penalty for violation in the fitness function

Implicit via the encoding of the problem

Implement a repair capability in case of an infeasible / unsound

For some problems, getting closer to the solution

In such a case, the fitness function needs to be

‘C’, ‘A’, ’T' ‘CAT’