SlideShare une entreprise Scribd logo
1  sur  21
Télécharger pour lire hors ligne
Fitness Inheritance in BOA
     Martin Pelikan                       Kumara Sastry
   Dept. of Math and CS                       Illinois GA Lab
Univ. of Missouri at St. Louis   Univ. of Illinois at Urbana-Champaign




                                                                  1
Motivation
Bayesian optimization algorithm (BOA)
  Scales up on decomposable problems
   O(n)-O(n2) evaluations until convergence
Expensive evaluations
  Real-world evaluations can be complex
  FEA, simulation, …
  O(n2) is often not enough
This paper
  Extend probabilistic model to include fitness info
  Use model to evaluate part of the population


                                                       2
Outline
BOA basics
Fitness inheritance in BOA
  Extend Bayesian networks with fitness.
  Use extended model for evaluation.
Experiments
Future work
Summary and conclusions


                                           3
Bayesian Optimization Alg. (BOA)
Pelikan, Goldberg, and Cantu-Paz (1998)
Similar to genetic algorithms (GAs)
Replace mutation + crossover by
  Build Bayesian network to model selected solutions.
  Sample Bayesian network to generate new candidate
  solutions.




                                                        4
BOA
                             Bayesian          New
 Current                     network
                Selection                    population
population




         Restricted tournament replacement
                                                      5
Bayesian Networks (BNs)
2 components
  Structure
    directed acyclic graph
    nodes = variables (string positions)
    Edges = dependencies between variables
  Parameters
    Conditional probabilities p(X|Px), where
        X is a variable
        Px are parents of X (variables that X depends on)



                                                            6
BN example
                 A   B   p(A|B)
                 0   0    0.10
                 0   1    0.60
    A            1   0    0.90
                 1   1    0.40


B          C             C        A   p(C|A)
                         0        0    0.80
                         0        1    0.55
B       p(B)
                         1        0    0.20
0       0.25
                         1        1    0.45
1       0.75                               7
Extending BNs with fitness info
                                       A     B       p(A|B)    f(A|B)
Basic idea
                                       0     0        0.10       -0.5
   Don’t work only with
   conditional probabilities           0     1        0.60       0.5
   Add also fitness info for           1     0        0.90       0.3
   fitness estimation                  1     1        0.40       -0.3

Fitness info attached to p(X|Px) denoted by f(X|Px)
  Contribution of X restricted by Px
   f (X = x | Px = px ) = f (X = x, Px = px ) − f (Px = px )
   f (X = x, Px = px )
       avg. fitness of solutions with X=x and Px=px
   f (Px = px )
      avg. fitness of solutions with Px
                                                                        8
Estimating fitness
Equation

                                       (          )
                                  n
    f (X1 , X2 ,K , X n ) = favg + ∑ f Xi | PXi
                                 i=0



In words
  Fitness = avg. fitness + avg. contribution of each bit
  Avg. contributions taken w.r.t. context from BN


                                                       9
BNs with decision trees
Local structures in BNs
  More efficient representation for p ( X | Px )
Example for p ( A | B C )
                                B
                         0              1

                p( A | B = 0)                C
                                    0               1


                     p( A | B = 1, C = 0 )       p( A | B = 1, C = 1)
                                                                        10
BNs with decision trees + fitness
Same idea
  Attach fitness info to
  each probability
                                 B
                           0             1

                p( A | B = 0)                C
                f ( A | B = 0)       0           1


                      p ( A | B = 1, C = 0) p ( A | B = 1, C = 1)
                      f ( A | B = 1, C = 0) f ( A | B = 1, C = 1)
                                                                    11
Estimating fitness again
Same as before…because both BNs represent the same
Equation

                                               (           )
                                         n
       f ( X 1 , X 2 ,K , X n ) = f avg + ∑ f X i | PX i
                                        i =0


In words
  Fitness = avg. fitness + avg. contribution of each bit
  Avg. contributions taken w.r.t. context from BN


                                                               12
Where to learn fitness from?
Evaluate entire initial population
Choose inheritance proportion, pi
After that
  Evaluate (1-pi) proportion of offspring
  Use evaluated parents + evaluated offspring to learn
  Estimate fitness of the remaining prop. pi
  Sample for learning: N(1-pi) to N+N(1-pi)
     Often, 2N(1-pi)


                                                     13
Simple example: Onemax
Onemax
                                        n
              f ( X 1 , X 2 ,K , X n ) = ∑ X i
                                       i =1

What happens?
  Average fitness grows (as predicted by theory)
  No context is necessary
  Fitness contributions stay constant
              f ( X i = 1) = +0.5
              f ( X i = 0 ) = −0.5
                                                   14
Experiments
Problems
  50-bit onemax
  10 traps of order 4
  10 traps of order 5
Settings
  Inheritance proportion from 0 to 0.999
  Minimum population size for reliable convergence
  Many runs for each setting (300 runs for each setting)
Output
  Speed-up (in terms of real fitness evaluations)


                                                           15
Onemax
                                   35
Speed-up (w.r.t. no inheritance)


                                   30

                                   25

                                   20

                                   15

                                   10

                                    5

                                    0
                                        0   0.2    0.4    0.6        0.8   1
                                              Proportion inherited
                                                                               16
Trap-4
                                   35
Speed-up (w.r.t. no inheritance)

                                   30

                                   25

                                   20

                                   15

                                   10

                                    5

                                    0
                                        0   0.2    0.4    0.6    0.8   1
                                              Proportion inherited
                                                                           17
Trap-5
                                   60
Speed-up (w.r.t. no inheritance)


                                   50

                                   40

                                   30

                                   20

                                   10

                                    0
                                        0   0.2    0.4    0.6        0.8   1
                                              Proportion inherited
                                                                               18
Discussion
Inheritance proportion
  High proportions of inheritance work great.
Speed-up
  Optimal speed-up of 30-53
  High speed-up for almost any setting
  The tougher the problem, the better the speed-up
Why so good?
  Learning probabilistic model difficult, so accurate
  fitness info can be added at not much extra cost.

                                                        19
Conclusions
Fitness inheritance works great in BOA
Theory now exists (Satry et al., 2004) that
explains these results
High proportions of inheritance lead to high
speed-ups
Challenging problems allow much speed-up
Useful for practitioners with computationally
complex fitness function

                                                20
Contact
Martin Pelikan
Dept. of Math and Computer Science, 320 CCB
University of Missouri at St. Louis
8001 Natural Bridge Rd.
St. Louis, MO 63121

E-mail: pelikan@cs.umsl.edu
WWW: http://www.cs.umsl.edu/~pelikan/

                                              21

Contenu connexe

Tendances

Principal component analysis and matrix factorizations for learning (part 2) ...
Principal component analysis and matrix factorizations for learning (part 2) ...Principal component analysis and matrix factorizations for learning (part 2) ...
Principal component analysis and matrix factorizations for learning (part 2) ...
zukun
 
MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...
MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...
MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...
The Statistical and Applied Mathematical Sciences Institute
 
Rank awarealgs small11
Rank awarealgs small11Rank awarealgs small11
Rank awarealgs small11
Jules Esp
 
CVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future Trend
CVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future TrendCVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future Trend
CVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future Trend
zukun
 
05 history of cv a machine learning (theory) perspective on computer vision
05  history of cv a machine learning (theory) perspective on computer vision05  history of cv a machine learning (theory) perspective on computer vision
05 history of cv a machine learning (theory) perspective on computer vision
zukun
 
Bayesian inversion of deterministic dynamic causal models
Bayesian inversion of deterministic dynamic causal modelsBayesian inversion of deterministic dynamic causal models
Bayesian inversion of deterministic dynamic causal models
khbrodersen
 
Physics of Algorithms Talk
Physics of Algorithms TalkPhysics of Algorithms Talk
Physics of Algorithms Talk
jasonj383
 

Tendances (20)

Session 6
Session 6Session 6
Session 6
 
Principal component analysis and matrix factorizations for learning (part 2) ...
Principal component analysis and matrix factorizations for learning (part 2) ...Principal component analysis and matrix factorizations for learning (part 2) ...
Principal component analysis and matrix factorizations for learning (part 2) ...
 
MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...
MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...
MUMS: Bayesian, Fiducial, and Frequentist Conference - Multiscale Analysis of...
 
Image denoising
Image denoisingImage denoising
Image denoising
 
Computational methods for Bayesian model choice
Computational methods for Bayesian model choiceComputational methods for Bayesian model choice
Computational methods for Bayesian model choice
 
Mesh Processing Course : Mesh Parameterization
Mesh Processing Course : Mesh ParameterizationMesh Processing Course : Mesh Parameterization
Mesh Processing Course : Mesh Parameterization
 
Rank awarealgs small11
Rank awarealgs small11Rank awarealgs small11
Rank awarealgs small11
 
CVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future Trend
CVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future TrendCVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future Trend
CVPR2010: Advanced ITinCVPR in a Nutshell: part 7: Future Trend
 
The multilayer perceptron
The multilayer perceptronThe multilayer perceptron
The multilayer perceptron
 
Alternating direction
Alternating directionAlternating direction
Alternating direction
 
05 history of cv a machine learning (theory) perspective on computer vision
05  history of cv a machine learning (theory) perspective on computer vision05  history of cv a machine learning (theory) perspective on computer vision
05 history of cv a machine learning (theory) perspective on computer vision
 
Bayesian inversion of deterministic dynamic causal models
Bayesian inversion of deterministic dynamic causal modelsBayesian inversion of deterministic dynamic causal models
Bayesian inversion of deterministic dynamic causal models
 
Physics of Algorithms Talk
Physics of Algorithms TalkPhysics of Algorithms Talk
Physics of Algorithms Talk
 
Maximum likelihood estimation of regularisation parameters in inverse problem...
Maximum likelihood estimation of regularisation parameters in inverse problem...Maximum likelihood estimation of regularisation parameters in inverse problem...
Maximum likelihood estimation of regularisation parameters in inverse problem...
 
Iwsmbvs
IwsmbvsIwsmbvs
Iwsmbvs
 
Semi-Supervised Regression using Cluster Ensemble
Semi-Supervised Regression using Cluster EnsembleSemi-Supervised Regression using Cluster Ensemble
Semi-Supervised Regression using Cluster Ensemble
 
Asymptotics for discrete random measures
Asymptotics for discrete random measuresAsymptotics for discrete random measures
Asymptotics for discrete random measures
 
Approximate Bayesian Computation on GPUs
Approximate Bayesian Computation on GPUsApproximate Bayesian Computation on GPUs
Approximate Bayesian Computation on GPUs
 
Linear Bayesian update surrogate for updating PCE coefficients
Linear Bayesian update surrogate for updating PCE coefficientsLinear Bayesian update surrogate for updating PCE coefficients
Linear Bayesian update surrogate for updating PCE coefficients
 
Macrocanonical models for texture synthesis
Macrocanonical models for texture synthesisMacrocanonical models for texture synthesis
Macrocanonical models for texture synthesis
 

En vedette

Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...
Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...
Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...
Martin Pelikan
 
Efficiency Enhancement of Estimation of Distribution Algorithms
Efficiency Enhancement of Estimation of Distribution AlgorithmsEfficiency Enhancement of Estimation of Distribution Algorithms
Efficiency Enhancement of Estimation of Distribution Algorithms
Martin Pelikan
 
Initial-Population Bias in the Univariate Estimation of Distribution Algorithm
Initial-Population Bias in the Univariate Estimation of Distribution AlgorithmInitial-Population Bias in the Univariate Estimation of Distribution Algorithm
Initial-Population Bias in the Univariate Estimation of Distribution Algorithm
Martin Pelikan
 
Estimation of Distribution Algorithms Tutorial
Estimation of Distribution Algorithms TutorialEstimation of Distribution Algorithms Tutorial
Estimation of Distribution Algorithms Tutorial
Martin Pelikan
 

En vedette (17)

Effects of a Deterministic Hill climber on hBOA
Effects of a Deterministic Hill climber on hBOAEffects of a Deterministic Hill climber on hBOA
Effects of a Deterministic Hill climber on hBOA
 
Intelligent Bias of Network Structures in the Hierarchical BOA
Intelligent Bias of Network Structures in the Hierarchical BOAIntelligent Bias of Network Structures in the Hierarchical BOA
Intelligent Bias of Network Structures in the Hierarchical BOA
 
Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...
Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...
Analysis of Evolutionary Algorithms on the One-Dimensional Spin Glass with Po...
 
Empirical Analysis of ideal recombination on random decomposable problems
Empirical Analysis of ideal recombination on random decomposable problemsEmpirical Analysis of ideal recombination on random decomposable problems
Empirical Analysis of ideal recombination on random decomposable problems
 
Efficiency Enhancement of Estimation of Distribution Algorithms
Efficiency Enhancement of Estimation of Distribution AlgorithmsEfficiency Enhancement of Estimation of Distribution Algorithms
Efficiency Enhancement of Estimation of Distribution Algorithms
 
Towards billion bit optimization via parallel estimation of distribution algo...
Towards billion bit optimization via parallel estimation of distribution algo...Towards billion bit optimization via parallel estimation of distribution algo...
Towards billion bit optimization via parallel estimation of distribution algo...
 
iBOA: The Incremental Bayesian Optimization Algorithm
iBOA: The Incremental Bayesian Optimization AlgorithmiBOA: The Incremental Bayesian Optimization Algorithm
iBOA: The Incremental Bayesian Optimization Algorithm
 
Simplified Runtime Analysis of Estimation of Distribution Algorithms
Simplified Runtime Analysis of Estimation of Distribution AlgorithmsSimplified Runtime Analysis of Estimation of Distribution Algorithms
Simplified Runtime Analysis of Estimation of Distribution Algorithms
 
Spurious Dependencies and EDA Scalability
Spurious Dependencies and EDA ScalabilitySpurious Dependencies and EDA Scalability
Spurious Dependencies and EDA Scalability
 
Using Previous Models to Bias Structural Learning in the Hierarchical BOA
Using Previous Models to Bias Structural Learning in the Hierarchical BOAUsing Previous Models to Bias Structural Learning in the Hierarchical BOA
Using Previous Models to Bias Structural Learning in the Hierarchical BOA
 
Initial-Population Bias in the Univariate Estimation of Distribution Algorithm
Initial-Population Bias in the Univariate Estimation of Distribution AlgorithmInitial-Population Bias in the Univariate Estimation of Distribution Algorithm
Initial-Population Bias in the Univariate Estimation of Distribution Algorithm
 
Transfer Learning, Soft Distance-Based Bias, and the Hierarchical BOA
Transfer Learning, Soft Distance-Based Bias, and the Hierarchical BOATransfer Learning, Soft Distance-Based Bias, and the Hierarchical BOA
Transfer Learning, Soft Distance-Based Bias, and the Hierarchical BOA
 
Using Problem-Specific Knowledge and Learning from Experience in Estimation o...
Using Problem-Specific Knowledge and Learning from Experience in Estimation o...Using Problem-Specific Knowledge and Learning from Experience in Estimation o...
Using Problem-Specific Knowledge and Learning from Experience in Estimation o...
 
Analyzing Probabilistic Models in Hierarchical BOA on Traps and Spin Glasses
Analyzing Probabilistic Models in Hierarchical BOA on Traps and Spin GlassesAnalyzing Probabilistic Models in Hierarchical BOA on Traps and Spin Glasses
Analyzing Probabilistic Models in Hierarchical BOA on Traps and Spin Glasses
 
The Bayesian Optimization Algorithm with Substructural Local Search
The Bayesian Optimization Algorithm with Substructural Local SearchThe Bayesian Optimization Algorithm with Substructural Local Search
The Bayesian Optimization Algorithm with Substructural Local Search
 
Order Or Not: Does Parallelization of Model Building in hBOA Affect Its Scala...
Order Or Not: Does Parallelization of Model Building in hBOA Affect Its Scala...Order Or Not: Does Parallelization of Model Building in hBOA Affect Its Scala...
Order Or Not: Does Parallelization of Model Building in hBOA Affect Its Scala...
 
Estimation of Distribution Algorithms Tutorial
Estimation of Distribution Algorithms TutorialEstimation of Distribution Algorithms Tutorial
Estimation of Distribution Algorithms Tutorial
 

Similaire à Fitness inheritance in the Bayesian optimization algorithm

GonzalezGinestetResearchDay2016
GonzalezGinestetResearchDay2016GonzalezGinestetResearchDay2016
GonzalezGinestetResearchDay2016
Pablo Ginestet
 
Jylee probabilistic reasoning with bayesian networks
Jylee probabilistic reasoning with bayesian networksJylee probabilistic reasoning with bayesian networks
Jylee probabilistic reasoning with bayesian networks
Jungyeol
 

Similaire à Fitness inheritance in the Bayesian optimization algorithm (20)

Unit IV UNCERTAINITY AND STATISTICAL REASONING in AI K.Sundar,AP/CSE,VEC
Unit IV UNCERTAINITY AND STATISTICAL REASONING in AI K.Sundar,AP/CSE,VECUnit IV UNCERTAINITY AND STATISTICAL REASONING in AI K.Sundar,AP/CSE,VEC
Unit IV UNCERTAINITY AND STATISTICAL REASONING in AI K.Sundar,AP/CSE,VEC
 
Module 4 - Linear Model for Classification.pptx
Module 4 - Linear Model for Classification.pptxModule 4 - Linear Model for Classification.pptx
Module 4 - Linear Model for Classification.pptx
 
Probability cheatsheet
Probability cheatsheetProbability cheatsheet
Probability cheatsheet
 
Probability Cheatsheet.pdf
Probability Cheatsheet.pdfProbability Cheatsheet.pdf
Probability Cheatsheet.pdf
 
Probability cheatsheet
Probability cheatsheetProbability cheatsheet
Probability cheatsheet
 
An overview of Bayesian testing
An overview of Bayesian testingAn overview of Bayesian testing
An overview of Bayesian testing
 
Bayes 6
Bayes 6Bayes 6
Bayes 6
 
ISBA 2016: Foundations
ISBA 2016: FoundationsISBA 2016: Foundations
ISBA 2016: Foundations
 
Statistics symposium talk, Harvard University
Statistics symposium talk, Harvard UniversityStatistics symposium talk, Harvard University
Statistics symposium talk, Harvard University
 
Bayes Classification
Bayes ClassificationBayes Classification
Bayes Classification
 
GonzalezGinestetResearchDay2016
GonzalezGinestetResearchDay2016GonzalezGinestetResearchDay2016
GonzalezGinestetResearchDay2016
 
AIML unit-2(1).ppt
AIML unit-2(1).pptAIML unit-2(1).ppt
AIML unit-2(1).ppt
 
Jylee probabilistic reasoning with bayesian networks
Jylee probabilistic reasoning with bayesian networksJylee probabilistic reasoning with bayesian networks
Jylee probabilistic reasoning with bayesian networks
 
Unit-2 Bayes Decision Theory.pptx
Unit-2 Bayes Decision Theory.pptxUnit-2 Bayes Decision Theory.pptx
Unit-2 Bayes Decision Theory.pptx
 
IT for Business Intelligence Term Paper
IT for Business Intelligence Term PaperIT for Business Intelligence Term Paper
IT for Business Intelligence Term Paper
 
Columbia workshop [ABC model choice]
Columbia workshop [ABC model choice]Columbia workshop [ABC model choice]
Columbia workshop [ABC model choice]
 
Probability based learning (in book: Machine learning for predictve data anal...
Probability based learning (in book: Machine learning for predictve data anal...Probability based learning (in book: Machine learning for predictve data anal...
Probability based learning (in book: Machine learning for predictve data anal...
 
ch8Bayes.pptx
ch8Bayes.pptxch8Bayes.pptx
ch8Bayes.pptx
 
ch8Bayes.ppt
ch8Bayes.pptch8Bayes.ppt
ch8Bayes.ppt
 
ch8Bayes.ppt
ch8Bayes.pptch8Bayes.ppt
ch8Bayes.ppt
 

Plus de Martin Pelikan

Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...
Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...
Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...
Martin Pelikan
 
Computational complexity and simulation of rare events of Ising spin glasses
Computational complexity and simulation of rare events of Ising spin glasses Computational complexity and simulation of rare events of Ising spin glasses
Computational complexity and simulation of rare events of Ising spin glasses
Martin Pelikan
 

Plus de Martin Pelikan (8)

Population Dynamics in Conway’s Game of Life and its Variants
Population Dynamics in Conway’s Game of Life and its VariantsPopulation Dynamics in Conway’s Game of Life and its Variants
Population Dynamics in Conway’s Game of Life and its Variants
 
Image segmentation using a genetic algorithm and hierarchical local search
Image segmentation using a genetic algorithm and hierarchical local searchImage segmentation using a genetic algorithm and hierarchical local search
Image segmentation using a genetic algorithm and hierarchical local search
 
Distance-based bias in model-directed optimization of additively decomposable...
Distance-based bias in model-directed optimization of additively decomposable...Distance-based bias in model-directed optimization of additively decomposable...
Distance-based bias in model-directed optimization of additively decomposable...
 
Pairwise and Problem-Specific Distance Metrics in the Linkage Tree Genetic Al...
Pairwise and Problem-Specific Distance Metrics in the Linkage Tree Genetic Al...Pairwise and Problem-Specific Distance Metrics in the Linkage Tree Genetic Al...
Pairwise and Problem-Specific Distance Metrics in the Linkage Tree Genetic Al...
 
Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...
Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...
Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor...
 
Finding Ground States of Sherrington-Kirkpatrick Spin Glasses with Hierarchic...
Finding Ground States of Sherrington-Kirkpatrick Spin Glasses with Hierarchic...Finding Ground States of Sherrington-Kirkpatrick Spin Glasses with Hierarchic...
Finding Ground States of Sherrington-Kirkpatrick Spin Glasses with Hierarchic...
 
Computational complexity and simulation of rare events of Ising spin glasses
Computational complexity and simulation of rare events of Ising spin glasses Computational complexity and simulation of rare events of Ising spin glasses
Computational complexity and simulation of rare events of Ising spin glasses
 
Hybrid Evolutionary Algorithms on Minimum Vertex Cover for Random Graphs
Hybrid Evolutionary Algorithms on Minimum Vertex Cover for Random GraphsHybrid Evolutionary Algorithms on Minimum Vertex Cover for Random Graphs
Hybrid Evolutionary Algorithms on Minimum Vertex Cover for Random Graphs
 

Dernier

一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书
一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书
一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书
atedyxc
 
Zepto Case study(On Track to Profitability).pptx
Zepto Case study(On Track to Profitability).pptxZepto Case study(On Track to Profitability).pptx
Zepto Case study(On Track to Profitability).pptx
aryan963438
 
一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书
一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书
一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书
atedyxc
 
NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...
NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...
NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...
Amil baba
 
PPT- Chapter 5- Cost Allocation presentation pdf
PPT- Chapter 5- Cost Allocation presentation pdfPPT- Chapter 5- Cost Allocation presentation pdf
PPT- Chapter 5- Cost Allocation presentation pdf
Kalkaye
 
State Space Tutorial.pptxjjjjjjjjjjjjjjj
State Space Tutorial.pptxjjjjjjjjjjjjjjjState Space Tutorial.pptxjjjjjjjjjjjjjjj
State Space Tutorial.pptxjjjjjjjjjjjjjjj
joshuaclack73
 

Dernier (20)

What exchange can I sell my pi coins in 2024
What exchange can I sell my pi coins in 2024What exchange can I sell my pi coins in 2024
What exchange can I sell my pi coins in 2024
 
Indirect tax .pptx Supply under GST, Charges of GST
Indirect tax .pptx  Supply under GST, Charges of GSTIndirect tax .pptx  Supply under GST, Charges of GST
Indirect tax .pptx Supply under GST, Charges of GST
 
Human Capital: Education and Health in Economic Development
Human Capital:  Education and Health      in Economic DevelopmentHuman Capital:  Education and Health      in Economic Development
Human Capital: Education and Health in Economic Development
 
一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书
一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书
一比一原版(BU毕业证书)波士顿大学毕业证成绩单学位证书
 
Consumer rights and its importance and value
Consumer rights and its importance and valueConsumer rights and its importance and value
Consumer rights and its importance and value
 
Zepto Case study(On Track to Profitability).pptx
Zepto Case study(On Track to Profitability).pptxZepto Case study(On Track to Profitability).pptx
Zepto Case study(On Track to Profitability).pptx
 
is pi Network coin available for sale in 2024
is pi Network coin available for sale in 2024is pi Network coin available for sale in 2024
is pi Network coin available for sale in 2024
 
Monthly Market Risk Update: May 2024 [SlideShare]
Monthly Market Risk Update: May 2024 [SlideShare]Monthly Market Risk Update: May 2024 [SlideShare]
Monthly Market Risk Update: May 2024 [SlideShare]
 
How can I withdraw my pi coins to real money in India.
How can I withdraw my pi coins to real money in India.How can I withdraw my pi coins to real money in India.
How can I withdraw my pi coins to real money in India.
 
一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书
一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书
一比一原版(KPU毕业证书)昆特兰理工大学毕业证成绩单学位证书
 
how do i sell pi coins in Pakistan at the best rate.
how do i sell pi coins in Pakistan at the best rate.how do i sell pi coins in Pakistan at the best rate.
how do i sell pi coins in Pakistan at the best rate.
 
What is an ecosystem in crypto .pdf
What  is  an  ecosystem  in  crypto .pdfWhat  is  an  ecosystem  in  crypto .pdf
What is an ecosystem in crypto .pdf
 
NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...
NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...
NO1 Best kala jadu karne wale ka contact number kala jadu karne wale baba kal...
 
International economics – 2 classical theories of IT
International economics – 2 classical theories of ITInternational economics – 2 classical theories of IT
International economics – 2 classical theories of IT
 
Maximize Your Business Potential with Falcon Invoice Discounting
Maximize Your Business Potential with Falcon Invoice DiscountingMaximize Your Business Potential with Falcon Invoice Discounting
Maximize Your Business Potential with Falcon Invoice Discounting
 
Fintech Belgium General Assembly and Anniversary Event 2024
Fintech Belgium General Assembly and Anniversary Event 2024Fintech Belgium General Assembly and Anniversary Event 2024
Fintech Belgium General Assembly and Anniversary Event 2024
 
Top 5 Asset Baked Tokens (ABT) to Invest in the Year 2024.pdf
Top 5 Asset Baked Tokens (ABT) to Invest in the Year 2024.pdfTop 5 Asset Baked Tokens (ABT) to Invest in the Year 2024.pdf
Top 5 Asset Baked Tokens (ABT) to Invest in the Year 2024.pdf
 
PPT- Chapter 5- Cost Allocation presentation pdf
PPT- Chapter 5- Cost Allocation presentation pdfPPT- Chapter 5- Cost Allocation presentation pdf
PPT- Chapter 5- Cost Allocation presentation pdf
 
Abhay Bhutada: A Journey of Transformation and Leadership
Abhay Bhutada: A Journey of Transformation and LeadershipAbhay Bhutada: A Journey of Transformation and Leadership
Abhay Bhutada: A Journey of Transformation and Leadership
 
State Space Tutorial.pptxjjjjjjjjjjjjjjj
State Space Tutorial.pptxjjjjjjjjjjjjjjjState Space Tutorial.pptxjjjjjjjjjjjjjjj
State Space Tutorial.pptxjjjjjjjjjjjjjjj
 

Fitness inheritance in the Bayesian optimization algorithm

  • 1. Fitness Inheritance in BOA Martin Pelikan Kumara Sastry Dept. of Math and CS Illinois GA Lab Univ. of Missouri at St. Louis Univ. of Illinois at Urbana-Champaign 1
  • 2. Motivation Bayesian optimization algorithm (BOA) Scales up on decomposable problems O(n)-O(n2) evaluations until convergence Expensive evaluations Real-world evaluations can be complex FEA, simulation, … O(n2) is often not enough This paper Extend probabilistic model to include fitness info Use model to evaluate part of the population 2
  • 3. Outline BOA basics Fitness inheritance in BOA Extend Bayesian networks with fitness. Use extended model for evaluation. Experiments Future work Summary and conclusions 3
  • 4. Bayesian Optimization Alg. (BOA) Pelikan, Goldberg, and Cantu-Paz (1998) Similar to genetic algorithms (GAs) Replace mutation + crossover by Build Bayesian network to model selected solutions. Sample Bayesian network to generate new candidate solutions. 4
  • 5. BOA Bayesian New Current network Selection population population Restricted tournament replacement 5
  • 6. Bayesian Networks (BNs) 2 components Structure directed acyclic graph nodes = variables (string positions) Edges = dependencies between variables Parameters Conditional probabilities p(X|Px), where X is a variable Px are parents of X (variables that X depends on) 6
  • 7. BN example A B p(A|B) 0 0 0.10 0 1 0.60 A 1 0 0.90 1 1 0.40 B C C A p(C|A) 0 0 0.80 0 1 0.55 B p(B) 1 0 0.20 0 0.25 1 1 0.45 1 0.75 7
  • 8. Extending BNs with fitness info A B p(A|B) f(A|B) Basic idea 0 0 0.10 -0.5 Don’t work only with conditional probabilities 0 1 0.60 0.5 Add also fitness info for 1 0 0.90 0.3 fitness estimation 1 1 0.40 -0.3 Fitness info attached to p(X|Px) denoted by f(X|Px) Contribution of X restricted by Px f (X = x | Px = px ) = f (X = x, Px = px ) − f (Px = px ) f (X = x, Px = px ) avg. fitness of solutions with X=x and Px=px f (Px = px ) avg. fitness of solutions with Px 8
  • 9. Estimating fitness Equation ( ) n f (X1 , X2 ,K , X n ) = favg + ∑ f Xi | PXi i=0 In words Fitness = avg. fitness + avg. contribution of each bit Avg. contributions taken w.r.t. context from BN 9
  • 10. BNs with decision trees Local structures in BNs More efficient representation for p ( X | Px ) Example for p ( A | B C ) B 0 1 p( A | B = 0) C 0 1 p( A | B = 1, C = 0 ) p( A | B = 1, C = 1) 10
  • 11. BNs with decision trees + fitness Same idea Attach fitness info to each probability B 0 1 p( A | B = 0) C f ( A | B = 0) 0 1 p ( A | B = 1, C = 0) p ( A | B = 1, C = 1) f ( A | B = 1, C = 0) f ( A | B = 1, C = 1) 11
  • 12. Estimating fitness again Same as before…because both BNs represent the same Equation ( ) n f ( X 1 , X 2 ,K , X n ) = f avg + ∑ f X i | PX i i =0 In words Fitness = avg. fitness + avg. contribution of each bit Avg. contributions taken w.r.t. context from BN 12
  • 13. Where to learn fitness from? Evaluate entire initial population Choose inheritance proportion, pi After that Evaluate (1-pi) proportion of offspring Use evaluated parents + evaluated offspring to learn Estimate fitness of the remaining prop. pi Sample for learning: N(1-pi) to N+N(1-pi) Often, 2N(1-pi) 13
  • 14. Simple example: Onemax Onemax n f ( X 1 , X 2 ,K , X n ) = ∑ X i i =1 What happens? Average fitness grows (as predicted by theory) No context is necessary Fitness contributions stay constant f ( X i = 1) = +0.5 f ( X i = 0 ) = −0.5 14
  • 15. Experiments Problems 50-bit onemax 10 traps of order 4 10 traps of order 5 Settings Inheritance proportion from 0 to 0.999 Minimum population size for reliable convergence Many runs for each setting (300 runs for each setting) Output Speed-up (in terms of real fitness evaluations) 15
  • 16. Onemax 35 Speed-up (w.r.t. no inheritance) 30 25 20 15 10 5 0 0 0.2 0.4 0.6 0.8 1 Proportion inherited 16
  • 17. Trap-4 35 Speed-up (w.r.t. no inheritance) 30 25 20 15 10 5 0 0 0.2 0.4 0.6 0.8 1 Proportion inherited 17
  • 18. Trap-5 60 Speed-up (w.r.t. no inheritance) 50 40 30 20 10 0 0 0.2 0.4 0.6 0.8 1 Proportion inherited 18
  • 19. Discussion Inheritance proportion High proportions of inheritance work great. Speed-up Optimal speed-up of 30-53 High speed-up for almost any setting The tougher the problem, the better the speed-up Why so good? Learning probabilistic model difficult, so accurate fitness info can be added at not much extra cost. 19
  • 20. Conclusions Fitness inheritance works great in BOA Theory now exists (Satry et al., 2004) that explains these results High proportions of inheritance lead to high speed-ups Challenging problems allow much speed-up Useful for practitioners with computationally complex fitness function 20
  • 21. Contact Martin Pelikan Dept. of Math and Computer Science, 320 CCB University of Missouri at St. Louis 8001 Natural Bridge Rd. St. Louis, MO 63121 E-mail: pelikan@cs.umsl.edu WWW: http://www.cs.umsl.edu/~pelikan/ 21