static vs. dynamic populations in gas for coloring a dynamic graph
DESCRIPTION
Static vs. Dynamic Populations in GAs for Coloring a Dynamic Graph. GECCO ’14 July 16, 2014. Cara Monical [email protected] Forrest Stonedahl [email protected]. Imagine You Want To…. Dynamic Graph Coloring. Big Question. [Jin & Branke ‘05]. Genetic Algorithm. 4. 9. - PowerPoint PPT PresentationTRANSCRIPT
Static vs. Dynamic Populations in GAs for
Coloring a Dynamic Graph
Cara [email protected]
Forrest [email protected]
GECCO ’14July 16, 2014
Imagine You Want To…
Allocate
Registers
Frequencies
Batches
For Conflicti
ngVariables
Devices
Jobs
InInterpreted
ProgramMobile Ad
Hoc Network
Management System
Dynamic Graph ColoringGAs for Static Graph Coloring• Galinier & Hao ‘99• many othersOnline Static Graph Coloring• Lovász et. al. ‘89
Ant-Based Dynamic Graph Coloring• Preuveneers & Berbers ‘04
Big QuestionDynami
c Problem
Genetic Algorith
m
[Jin & Branke ‘05]
Genetic Algorithm
Population of
solutionsEvaluate fitness
Select fit individua
ls
Perform Crossov
er
Perform Mutatio
n
4
35
6
7
2
9
8
10
1
1
2
5
8
10
7
3
6
9 4Pop Size: 100
Greedy Decoder
Tournament, size 3
Evaluate fitness
Select fit individua
ls
Reproduction: OX1 & SWAP
Parent 1
Parent 2
Offspring
Population of
solutions
Perform Crossov
er
Perform Mutatio
n
AfterBefore
21 43 5 6 7 8 9 10
21 43 56 78 910
2 89 10
* *
143 5 6 7
298101
43 5 6 7
21 3 5 6 8 9 103 5 6 74
Rate: 70% Rate: 50%
[Starkweather ‘91]
Experimental SetupCD E A B
1. Graph 3. Dynamic Population (DGA)
2. DSATUR [Brélaz ‘79]
4. Static Population (SGA)
AE B D C
DC A E B
EC D B A
DA E B C
BA D C EE DCA B
AD B E C
EC D A B
0
0
0
CB D E A
AD B E C
CA E D B
AE B C D
4
3
EC D B A
DA B C E
DA E B C
AC D E B
3
4
4
3
3
3
3
3
4
3
A
E
DC
B
E DCA B
CB D E A
AD B E C
CA E D B
AE B C D
EC D B A
DA B C E
DA E B C
AC D E B
Experimental Setup1. Graph 3. Dynamic Population
(DGA)
2. DSATUR
4. Static Population (SGA)DE B F C
FC B E D
BD F E CB CDF E
DB C F E
0
0
3
CB D E A
AD B E C
CA E D B
AE B C D
4
3
7
6
CB F D E
BD E C F
FC E D B
BE C F D
EB F D C
FC E D B
BD E C F
BE C F D
6
FC E D B
DE C F B
BD F E C
BD C F E3
3
3
3
3
3
3
3
3
A
E
DC
B
F
B CDF E
EB F D C
FC E D B
BD E C F
BE C F D
FC E D B
DE C F B
BD F E C
BD C F E
Experimental ParametersGraph
Properties
DynamicPropertie
s
• n: Size, 100• p: Edge
density, .6• Structure
• G(n,p,cv) • Euclidean
• cv: Vertex change rate, .01
• e: Evolution a step, 1000
.05
.05
.05
.05
.025 .0167 .0125 .01
.1 .033 .2
.2
.075.15.025
.0375
.1
.03
.067 .04
(Some) Big Answers (For this Problem & Algorithm)
Dynamic Problem ≥ Succession of
Static Problems
1.Highly
Dynamic Problem ≈ Succession of
Static Problems
2.Slightly
Dynamic Problem > Succession of
Static Problems
3.
Thank YouCentre College Department of Computer Science and Department of MathematicsCentre College, John C. Young Program
Contact InformationCara Monical
University of Illinois at Urbana-Champaign Math Department
Forrest StonedahlAugustana College
CS and Math [email protected]
Performance vs. Edge Density
G(n,p,cv) Graphs Euclidean Graphs
Performance vs. EvolutionG(n,p,cv) Graphs Euclidean Graphs