diversity maintenance behavior on evolutionary multi-objective o ptimization

18
Diversity Maintenance Behavior on Evolutionary Multi-Objective Optimization Presenter : Tsung Yu Ho 2011.11.27 at TEILAB

Upload: sutton

Post on 23-Feb-2016

45 views

Category:

Documents


0 download

DESCRIPTION

Diversity Maintenance Behavior on Evolutionary Multi-Objective O ptimization . 2011.11.27 at TEILAB. Presenter : Tsung Yu Ho. Main Point of Today’s Presentation. Introduction multi-objective problems (MOP) Perato Front (non-dominated points) - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Diversity Maintenance Behavior on Evolutionary Multi-Objective

Optimization

Presenter : Tsung Yu Ho

2011.11.27 at TEILAB

Page 2: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Main Point of Today’s Presentation Introduction multi-objective problems (MOP)

Perato Front (non-dominated points)

Evolutionary multi-objective optimization (EMO) Perato dominance-based fitness

evaluation. Diversity maintenance Elitism

Diversity maintenance Want to observe diversity in high dimension

(D>4)

Page 3: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Related work Hisao Ishibuchi et.al. ,“A Many-Objective Test

Problem for Visually Examining Diversity Maintenance Behavior in a Decision Space”, GECCO 2011

A 2-D problems space is used for presenting many-objective problems.

Observer “diversity maintenance “ on current well-known EMO, such as NSGA-II, SPEA2

Page 4: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Multi-Objective Problems Perato Front (non-dominated points)

X

Y

Page 5: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Evolutionary multi-objective optimization (1) NSGA – II SPEA2

Fitness assignment

Density estimation

Y

X

Y

X

0

00

0

7

Page 6: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Evolutionary multi-objective optimization (2) SMS-EMOA

Hypervolume

Page 7: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

What’s the problems Observe diversity maintenance

2-D is clear thinking.

Manny-objective problems is hardly observed by using figure.

Need to design a test functions to evaluate diversity maintenance.

It is easy to observe if the problems is mapped to 2-D space.

Page 8: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

2-D distance minimization problems Buying a house nearest these location.

Convenience stores (Objective 1) MRT stations (Objective 2) School (Objective 3) Park (Objective 4)

Page 9: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

2-D Decision Space : Perato Front A simple example

A

B C

Perato Front

Page 10: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Adjust Problems Observe diversity

Page 11: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Experiments

Page 12: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Real world application The region is the range of perato front

Page 13: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Real World Perato Front The number of perato front in three part.

Page 14: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

What information that should be observed? Diversity maintenance

Number on difference region of Perato front

Small region of Perato front

Hypervolume

Page 15: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Experiment Results of distribution The number of solution in the smallest Pareto

region.

Page 16: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Experiment Results of diversity Observe with three points

Page 17: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Hypervolume The small value is worst.

The reference points : x {maximum value of each objective in perato front}

Page 18: Diversity  Maintenance Behavior  on  Evolutionary  Multi-Objective  O ptimization

Conclusion A 2-D problems space is used for presenting

many-objective problems. Observe well-known EMO.

The 2-D distance minimization problems. Adjust the region of Perato front Can be utilized in the real world application

The observation measurement Hypervolume Number on difference region of Perato front Small region of Perato front