Scribd
Upload
3 views

02_GeneticAlgorithms

The document discusses genetic algorithms (GAs) as a method for solving optimization problems through processes inspired by natural evolution. It covers key concepts such as representation of solutions, fitness estimation, selection methods, and genetic operators like crossover and mutation. Additionally, it highlights the limitations of GAs and their application in training neural networks, along with various types of encoding used in GAs.

Uploaded by

ursady4u
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

02_GeneticAlgorithms

The document discusses genetic algorithms (GAs) as a method for solving optimization problems through processes inspired by natural evolution. It covers key concepts such as representation of solutions, fitness estimation, selection methods, and genetic operators like crossover and mutation. Additionally, it highlights the limitations of GAs and their application in training neural networks, along with various types of encoding used in GAs.

Uploaded by

ursady4u
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 34

02 GENETIC ALGORITHMS

Spring 2023 CS6431 Natural Language Processing


Credits
1. B1: Machine learning: an algorithmic perspective. 2nd Edition,
Marsland, Stephen. CRC press, 2015
2. B2: Principles of Soft Computing. 3rd Edition. S. N. Sivanandam, S. N.
Deepa. Wiley, 2018.
3. https://www.tutorialspoint.com/genetic_algorithms/genetic_algorith
ms_parent_selection.htm
Assignment
Read:
B1: Chapter 10 (Till 10.3)

Problems:
B1:
Evolution as a search problem
 Competing animals and “Survival of the fittest”
 “Fittest” animals
◼ Livelonger
◼ Stronger
◼ More attractive

 Hence, they get more mates and produce more and “healthier” off springs
 Nature is biased towards
“fitter” animals for
reproduction
 Basisfor Genetic
Algorithms
 Parent chromosomes are
copied randomly to the
child
 Butcopy errors can
happen: Mutation
Genetic Algorithm (GA)
Modelling a problem as a GA
 A method for representing solutions as chromosomes (or

string of characters)
 A way to calculate the fitness of a solution
One generation
 A selection method to choose parents

 A way to generate offspring by breeding the parents

 A way to select next generation

Select, Produce, Repeat!


An Evolutionary Learning Example
 Knapsack Problem  Given a set of items, each with
a weight and a value,
determine the number of each
Size A1 Size A2
item to include in a collection so
that the total weight is less than
Size A4 or equal to a given limit and the
Pack Volume=b
Size A3 total value is as large as
possible.
Can we find k
objects which will fit Size A5
Size A6
the pack volume b
perfectly?
String Representation

𝐿=#items
𝑆= 1 1 0 0 0 1 0 1 0 1 0 0 0

𝑆𝑖 =1 if 𝑖 𝑡ℎ item is included in the solution, 0, otherwise


Fitness Estimation
 More valuable items => Better fitness
 Fitness Value = Value of all items in a solution
 Solution should be feasible
 Fitness Value = 0 (if items don’t fit)
Exploitation vs. Exploration
 Exploitation: use the best solution so far to explore further solutions
 Exploration: use sub-optimal solutions to explore further solutions
 Exploration: Give infeasible solutions a chance for next generation
 Fitness Value of infeasible solutions = Total value of all items -
2 ×value of extra items
Selecting Parents for the “Mating Pool”
 Exploitation: select the “fittest” solutions
 Exploration: allow some sub optimal solutions

1. Tournament Selection
2. Truncation Selection
3. Fitness Proportional Selection
Tournament Selection
 Way of selecting one parent (individual) at a time
 Choose k (the tournament size) individuals from the population at random
 Choose the best individual from the tournament with probability 𝑝

 Choose the second best individual with probability 𝑝 × (1 − 𝑝)


2
 Choose the third best individual with probability 𝑝 × ( 1 − 𝑝 )

…and so on (till one parent is selected)


 Deterministic version: 𝑝 = 1
Truncation Selection
 𝑀𝑃 = Pick 𝑓 fraction of the best strings in the mating pool
1
 Mating pool = × 𝑀𝑃
𝑓
 So that mating pool size = Initial population size
 Randomly shuffle the pool and make pairs
 Easy to implement but biased towards exploitation
Fitness Proportional Selection
 Select (with replacement) a string (𝛼) probabilistically in proportion to
its fitness
𝛼
𝐹
𝑝𝛼 =
σ𝛼′ 𝐹 𝛼′
 If 𝐹 can be –ve
𝛼
𝛼
exp(𝑠𝐹 )
𝑝 =
σ𝛼′ exp(𝑠𝐹 𝛼′ )
𝑠: selection strength
 higher 𝑠 gives higher (or less –ve) 𝐹 𝛼 a higher probability
Roulette Wheel Selection
Generating Offspring
 Genetic Operators
 CrossOver
 Mutation
Crossover

Single Point Multi Point Random

 Global exploration
 Offspring are radically different than their parents
Mutation
1
 Flip a bit with a low probability 𝑝 =
𝐿
Choosing Next Generation
 Choosing only off spring can be risky
 New generation can have lesser fitness values
 We can lose a really good sample

 Elitism
 Selecta few fittest strings from parents, say 𝑋
 Replace strings from offspring with 𝑋
◼ Replacement at random, or,
◼ Replace the least fit

 Tournament
 Two fittest parent and their offspring
 Tournament winners proceed to the next round
 Elitism and Tournament can lead to premature convergence
 Both promote fitter members
 After some time, same set of fittest members keep getting promoted

 Exploration is downplayed

 Solution:
 Niching (or “using island populations”)
 Fitness sharing
Niching
 Separate populations into subpopulations
 Each subpopulation converges independently to different local minima
 A few members of one sub-population are randomly injected as
“immigrants” to another.
Fitness Sharing
𝛼
𝐹
𝐹𝛼 =
#Times 𝛼 appears in a population
 Biased for uncommon strings
 But, can loose very good common strings
Modelling a problem as a GA
 A method for representing solutions as chromosomes (or string of
characters)
 A way to calculate the fitness of a solution

 Exploration vs. exploitation


 A selection method to choose parents One
 Tournament, Truncation, Fitness Proportional selection generation
 A way to generate offspring by breeding the parents
 Crossover and mutation
 A way to select next generation
 Elitism, Tournament, Niching, Fitness sharing

Select, Produce, Repeat! => Until stopping criteria is met


Four Color Theorem: Map Colouring Problem
 No more than four colors are required to color the regions of the map so
that no two adjacent regions have the same color.

 We will formulate three-Color-Problem using GA

Image Credit: https://brilliant.org/wiki/four-color-theorem/


Encoding Solutions
 Three colours: {black (𝑏), dark (𝑑), light(𝑙)}
 We assign a fixed order to region
 For a six-region map, a solution looks like
𝛼 = {𝑏𝑑𝑏𝑙𝑏𝑏}
Fitness Function
 A negative point for every two adjacent regions having the same color
 Fitness value can be negative
exp(𝑠𝐹 𝛼 )
σ𝛼′ exp(𝑠𝐹 𝛼′ )
Or,
𝐹 𝛼 = #Toal Boundaries − #Adjacent regions having same color
No negative fitness value
Genetic Operators

Mutation
Crossover
Limitations of GA
 Can get stuck to a local minima for a very long time
 Like a black box:
 We don’t know how error landscape looks like and how it is working
 No guarantee to converge
Training Neural Networks with GA
 Encode weights as strings
 Fitness function: sum-of-square errors
 Reasonably good results.

 Problems:
 Local error information at an o/p node is lost – the entire error is clubbed
into one number
 Not using gradient information
Types of Encoding in GA
 Binary Encoding: a string of 0s and 1s
 E.g.: Knapsack problem
 Permutation Encoding: string of numbers representing a sequence
 E.g.: Sorting a sequence of numbers
4 2 5 9 0 7 8 1 3 6
 How to do crossover?
4 2 5 9 0 7 8 1 3 6
+
8 1 9 4 2 5 6 3 0 7
=

4 2 5 9 0 8 1 6 3 7
 How to mutate?
 Value Encoding: a string of real numbers
 E.g.: Training weights in neural networks
 Tree Encoding: Each chromosome is a tree
 E.g.:Given input and output values, find a function, which will give the best
(closest to wanted) output to all inputs.

(+ x (/ 5 y))

You might also like