Genders and EAs Using Gestation Periods to Control Population Dynamics Cameron Johnson.

Genders and EAs

Using Gestation Periods to Control Population Dynamics

Cameron Johnson

Motivation & Justification• Inspiration from biology

• “Black Box” for EAs

Why Genders?

• Panmictic mating produces results

• Meta-EAs and self-adaptive, self-regulating EAs

Methods• Algorithm basics

– Fitness used as mate-selection algorithm– Gestation period

• Population size-control• Restriction on reproductive speed

– Child Support• Balance between own survival and offspring survival

• Behavioral Genes– Male and female child support %– Male and female faithfulness (expressed as %)– Male and female mutation rates (expressed as %)– Sex allele – male or female?

Mate Fitness• Females are simply ranked by normalized fitness

– The fittest female chooses her mate first

• Males’ fitness is modified from its base to create an “attractiveness”

Mate Selection & Child Support

• Females choose based on promises

• Male promise reduced for each promise made

• Male and female real fitnesses reduced by child support

Factors to Keep Track of

• Is the individual alive?

• Who are his parents (father & mother)?

• Is the individual pregnant?

• With whom did the individual last mate?

• How many children does the individual have?

4-Dimensional Spherical Test Function

Experimental Average: -4.5 Standard Deviation: 4.57

Standard Average: -.047Standard Deviation: .027

7-Dimensional Spherical Test Function

Experimental Average Fitness: -633.2Standard Deviation: 705.76

Standard Average Fitness: -.648Standard Deviation: .244

10-Dimensional Spherical Test Function

Experimental Average Fitness: -3946Standard Deviation: 6604.96

Standard Average Fitness: -2.8Standard Deviation: .64

Conclusions

• Performance is disappointing– Accuracy cannot keep up with standard algorithm

even on a simple problem

• Population cannot always recover from collapse due to premature convergence– Likely due to loss of genetic diversity

• Population dynamics are self-adaptive, so promise is shown, but not yet achieved

Future Work

• Rebuilding with a more efficient implementation for quicker data-taking

• Experiment with different mate-selection parameters for genetic diversity

• Try hard-set and heuristic-adjusted mutation rates

• Generally, continued analysis of causes for sub-optimal performance

Questions?

• “A man pushes a car up to a hotel and tells the owner he is bankrupt. Why?”

• “A man lies dead next to the rock that killed him. Why is his underwear visible?”

• “Fred and Gertrude lie dead amidst a puddle of water. Shards of broken glass are scattered everywhere. What killed them?”

• “Who is the greater inventor: Darwin for evolution, or Al Gore for the Internet?”

Answers!

• Now that would be telling, wouldn’t it?

4 Dimensions, First Run

0 0.5 1 1.5 2 2.5

x 104

0

1000

2000

3000

4000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-3

-2

-1

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

4 Dimensions, Second Run

4 Dimensions, Third Run

0 0.5 1 1.5 2 2.5

x 104

0

1000

2000

3000

4000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-3

-2

-1

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

4 Dimensions, Fourth Run

0 0.5 1 1.5 2 2.5

x 104

0

2000

4000

6000

8000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-3

-2

-1

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

4 Dimensions, Fifth Run

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

0

200

400

600Population vs. # of Fitness Evaluations

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

-3

-2

-1

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

4 Dimensions, Sixth Run

0 0.5 1 1.5 2 2.5

x 104

0

2000

4000

6000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-3

-2

-1

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

4 Dimensions, Seventh Run

4 Dimensions, Eighth Run

4 Dimensions, Ninth Run

4 Dimensions, Tenth Run

0 0.5 1 1.5 2 2.5

x 104

0

2000

4000

6000

8000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-3

-2

-1

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

7 Dimensions, First Run

7 Dimensions, 3rd and 4th Runs

0 0.5 1 1.5 2 2.5

x 104

0

500

1000

1500Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

7 Dimensions, 5th and 6th Runs

0 0.5 1 1.5 2 2.5

x 104

0

200

400

600Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

7 Dimensions, 7th and 8th Runs

7 Dimensions, 9th and 10th Runs

0 0.5 1 1.5 2 2.5

x 104

0

500

1000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

0 0.5 1 1.5 2 2.5

x 104

0

200

400

600Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

10 Dimensions, 1st and 2nd Runs

10 Dimensions, 3rd and 4th Runs

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

0

200

400

600Population vs. # of Fitness Evaluations

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

10 Dimensions, 5th and 6th Runs

10 Dimensions, 7th and 8th Runs

0 0.5 1 1.5 2 2.5

x 104

0

500

1000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

0

200

400

600Population vs. # of Fitness Evaluations

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

-8

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness

10 Dimensions, 9th and 10th Runs

0 0.5 1 1.5 2 2.5

x 104

0

1000

2000

3000Population vs. # of Fitness Evaluations

0 0.5 1 1.5 2 2.5

x 104

-6

-4

-2

0x 10

4 Fitness fs. # of Fitness Evaluations

best fitness

average fitness