25
Evolutionary Computation Biologically inspired algorithms BY: Andy Garrett YE Ziyu

Evolutionary Computation

  • Upload
    varen

  • View
    95

  • Download
    0

Embed Size (px)

DESCRIPTION

Evolutionary Computation. Biologically inspired algorithms. BY: Andy Garrett YE Ziyu. What is Evolutionary Computation. A subfield of artificial intelligence which mimics biology Used in optimization of black box problems Parallel processing. Types of Evolutionary Computation. - PowerPoint PPT Presentation

Citation preview

Page 1: Evolutionary Computation

Evolutionary Computation

Biologically inspired algorithms

BY:Andy GarrettYE Ziyu

Page 2: Evolutionary Computation

What is Evolutionary Computation

• A subfield of artificial intelligence which mimics biology• Used in optimization of black box problems• Parallel processing

Andy Garrett
Used to find the optimum solution to a problem where the relationship between input and ouput are not inherently clear
Andy Garrett
A large population is distributed to processors
Page 3: Evolutionary Computation

Types of Evolutionary Computation

• Evolutionary programing• Genetic algorithms• Evolutionary strategies• Genetic programing

• Genetic algorithms• Swarm intelligence

Andy Garrett
Genetic Algorithms use Genes and mutation
Andy Garrett
Particle swarm optimization (PSO) is a global optimization algorithm for dealing with problems in which a best solution can be represented as a point or surface in an n-dimensional space. Hypotheses are plotted in this space and seeded with an initial velocity, as well as a communication channel between the particles.[2][3] Particles then move through the solution space, and are evaluated according to some fitness criterion after each timestep. Over time, particles are accelerated towards those particles within their communication grouping which have better fitness values. The main advantage of such an approach over other global minimization strategies such as simulated annealing is that the large number of members that make up the particle swarm make the technique impressively resilient to the problem of local minima.
Andy Garrett
The four origanal catagories which were combined to create the field of Evolutionary Computation
Andy Garrett
We will focus on the two most prevalent of Evolutionary computational strategies of today
Page 4: Evolutionary Computation

Genetic Algorithms

Page 5: Evolutionary Computation

Genetic Algorithm——what is gene?

Biology:A certain DNA sequenceat a certain position of the chromosome.

Genetic Algorithm :A certain value of a certain element of the solution.

2 3 1 1 1

2 3 2 1 1

2 3 3 1 1

A certain element (an allele) of the solution (the chromosome)

Three alternative values (genes)

Page 6: Evolutionary Computation

Genetic Algorithm——what is gene?

Biology Genetic Algorithm

Genes

Chromosome

Fitness of a individualIn the environment

Genes

Solution

Performance of a solutionin the problem. (Fitness)

constitute constitute

determines determines

Page 7: Evolutionary Computation

Genetic Algorithm——what is gene?

In Genetic Algorithm, genes (values of elements of the solution) determine the fitness (performance) of a solution.

To solve a problem=

To find the combination of genes that provides the best fitness (performance)

Page 8: Evolutionary Computation

Genetic Algorithm——Initiation

To conduct evolution,We need a set of solutions.(A population)

Initially, the population is generated randomly. This isthe first generation. A two-dimension search space

dotted by randomly generated solutions(each solution consists of two elements,

x and y)

X

Y

Page 9: Evolutionary Computation

Genetic Algorithm——Reproduction: Crossover

Crossover is how we create new individuals from the existing ones.

2

3

1

4

4

1

2

2

1

3

Two solutions somehow be

selected as “parents”

2

3

1

4

4

1

2

2

1

3

Randomly selectone (or more)

point

2

3

2

1

3

1

2

1

4

4

Apply cross(Recombine the two solutions)

2

3

2

1

3

1

2

1

4

4

Finish!These will be two

Individuals inthe next generation

Page 10: Evolutionary Computation

Genetic Algorithm——Reproduction: Selection

• Individuals with higher fitness have a higher probability to be chosen as parents of the crossover operation.

• Survival of the fittest

Page 11: Evolutionary Computation

Genetic Algorithm——Reproduction: SelectionWhat’s the effect?

Genes associated with high fitness are more likely to be passed to the new generation.

After some generations, the average fitness of the population gets improved!

Page 12: Evolutionary Computation

Genetic Algorithm——Reproduction: Selection

In a graphic view: (use our two-dimension example)

The population gathers aroundthe optimal solution.

It’s like that the population is climbing the hill.

Problem solved?

X

Y

Page 13: Evolutionary Computation

Genetic Algorithm——Mutation

Problem: What if we have multiple hills in the searching

space?

The individuals may climb onto a hill that is not the highest.

Thus, they may gather around a local optimum.

Y

X

Y

(Local optimum)

(Global optimum)

Page 14: Evolutionary Computation

Genetic Algorithm——Mutation

According to the crossover operation, genes in the new generation only come from the previous generation.

Thus, once the solutions gather around a local optimum, they will be constrained in its vicinity!

They won’t find the global optimum.

X

Y

(Constraining region)

Page 15: Evolutionary Computation

Genetic Algorithm——Mutation

Mutation: Make random changes to some genes in each generation.

NEW genes are created!Solutions can jump out of the region.After some generations, they may probably gather around theglobal optimum.

X

Y

Page 16: Evolutionary Computation

Genetic Algorithm——Scenario

Step 1: Initiation (Randomly generate the first generation);Step 2: Mutation;Step 3: Fitness evaluation; Step 4: Reproduction:

Selection;Crossover;

Step 5: Go back to step 2, repeat this loop until a sufficiently good solution is found.

Page 17: Evolutionary Computation

Swarm intelligence

Page 18: Evolutionary Computation

Swarm Intelligence

Swarm intelligence=

cognition of individuals + communication

Application in optimization problems:Particle Swarm Optimization (PSO)

Page 19: Evolutionary Computation

Swarm Intelligence——Initiation

Randomly generate a set of solutions (called a swarm of particles),their initial positions,and their initial speeds.

X

Y

V2oV3o

V1o

Page 20: Evolutionary Computation

Swarm Intelligence——Travelling

Two forces are exertedon each particle:

X

Y

1. Force pointing to the bestsolution this particle has everpassed through (pbest)2. Force pointing to the bestsolution any particle has everpassed through (gbest) pbest gbest

pbest1

pbest2(gbest)

pbest3

Page 21: Evolutionary Computation

Swarm Intelligence——Travelling

Forces pointing to pbests:Fp1, Fp2, Fp3

These forces result from the cognition of individual particles.

X

Y

Fp3

Fp2

Fp1

Page 22: Evolutionary Computation

Swarm Intelligence——Travelling

Forces pointing to gbests:Fg1, Fg2, Fg3

These forces result from the communication among the particles.

X

Y

Fg3 Fg2Fg1

Page 23: Evolutionary Computation

Swarm Intelligence——Travelling

After some time, the particles would probablyfind some solutions thatare sufficiently close theglobal optimum.

X

Y

Fg3 Fg2Fg1Fp3

Fp2

Fp1

https://www.youtube.com/watch?v=j028fsZZZI4

Page 24: Evolutionary Computation

Evolutionary Computation

• Time complexity is not generally considered• Number of iterations required for convergence

Andy Garrett
Because of the unique nature of evolutionary computation the algorithm is not generally analyzied using time complexity. However the number of iterations needed for convergence is a good indication for the run time of the algorithm.
Page 25: Evolutionary Computation

Questions?