SlideShare a Scribd company logo
1 of 18
Download to read offline
Empirical Analysis of Ideal Crossover on
Random Additively Decomposable Problems

 Kumara Sastry1, Martin Pelikan2, David E. Goldberg1
        1Illinois  Genetic Algorithms Laboratory (IlliGAL)
  University of Illinois at Urbana-Champaign, Urbana, IL 61801
   2MissouriEstimation of Distribution Algorithm Lab (MEDAL)
        University of Missouri at St. Louis, St. Louis, MO
         ksastry@uiuc.edu, pelikan@cs.umsl.edu, deg@uiuc.edu
           http://www.illigal.uiuc.edu, http://medal.cs.umsl.edu




               Supported by AFOSR FA9550-06-1-0096, NSF DMR
               03-25939, and CAREER ECS-0547013. Computational
               results obtained using CSE’s Turing cluster.
Roadmap
Adversarial test problem design
Random additively decomposable problems
Ideal crossover
Scalability of selectorecombinative GAs
  Population sizing and Run duration
Experimental Procedure
Key Results
Summary and Conclusions




                                          2
Adversarial Test Problem Design
Test systems on boundary of design envelope
   Common approach in designing complex systems
GAs are complex systems [Goldberg, 2002]
GA design envelope characterized by different dimensions
of problem difficulty
   Thwart the mechanism of GAs to the extreme

                        Fluctuating



                                      R
                    P                      Noise
       Deception          Scaling


                                                           3
Random Additively Decomposable Problem
Focus on nearly decomposable problems [Simon, 1960]
Three desired features
   Scalability: Able to control problem size and difficulty
   Known optimum: Allows comparison of different solvers
   Easy problem instance generation

rADP fitness function:

   Si represents variable subset for ith subproblem
   Each subset consists of k bits
   gi is the fitness of the ith subproblem
   gi is sampled from uniform distribution U[0,1]


                                                              4
Ideal Crossover: Exchange Building Blocks

Population sizing and run duration models assume good
exchange of building blocks

Simulate what we ideally want to achieve with model-
building GAs
  For example, extended compact GA [Harik, 1999]
Ideal recombination operator
  Effectively exchange building blocks
  Don’t disrupt any building block

Uniform building-block-wise crossover
                                         BBs #1 and #3 exchanged
  Exchange BBs with probability 0.5


                                                                   5
Purpose: Analyze Ideal Crossover on rADPs

Analyze behavior of selectorecombinative GAs on rADPs
Verify the validity of lessons learned from adversarial test
problems
Expand the pool of test problems




                                                               6
Selectorecombinative GA Population Sizing

                                                   Noise-to-fitness
                                                   variance ratio

  Error tolerance                   # Components (# BBs)
     Signal-to-Noise ratio   # Competing sub-components


                                  Gambler’s ruin model
                                  [Harik, et al, 1997]
                                      Combines decision
                                      making and supply
                                      models
                                  Additive Gaussian noise
                                  with variance σ2N


                                                                      7
GA Run Duration (Selection)

    Selection-Intensity based model [Bulmer, 1980; Mühlenbein &
    Schlierkamp-Voosen, 1993; Thierens & Goldberg, 1994; Bäck, 1994; Miller & Goldberg,
    1995 & 1996]

        Derived for the OneMax problem
        Applicable to additively-separable problems [Miller, 1997]
                    Problem size (m·k )




                   Selection Intensity




[Miller & Goldberg, 1995;
Sastry & Goldberg, 2002]
                                                                                          8
GA Run Duration (Drift)

Accumulation of stochastic errors due to finite population

Proportion of competing sub-solutions change due to drift

Drift time [Goldberg & Segrest, 1987]:



Substituting population sizing bound




                                                             9
Signal-to-Noise Ratio for rADPs

Signal d is the fitness difference between best and second
best sub-solutions

jth order statistic follows a Beta distribution with
α = j and β = 2k-j+1

Probability density function (p.d.f) of d:

p.d.f. of sub-solution fitness variance approximation




E[1/d] = 2k and E[σ2BB] ≈ 1/12

                                                             10
Assumptions and Experimental Setup
Non-overlapping sub-problems
Identical sub-problems across different partitions
   g1 = g2 = … = gm

Selectorecombinative GA
   Binary tournament selection
10,000 random problem instances
   m = 5 – 50, k = 3, 4, and 5
Minimum population size determined by bisection method
   Population correctly converges to at least m-1 out of m BBs in
   49 out of 50 independent runs
   Averaged over 30 bisection runs
   Results averaged over 1,500 GA runs
                                                                    11
Population Sizing & Run Duration Histograms

            Population size                   Run duration
               m = 50                           m = 50




  Tail increases with m
  0.15-0.59% of rADP instances require # evals greater
  than 3σ from the median

                                                             12
Easy and Hard Problem Instances

                         Hard instance
Subsolution fitness




                                   Min signal
                                   Max noise



                      Sorted subsolution index

                         Easy instance
Subsolution fitness




                          Max signal
                          Min noise

                      Sorted subsolution index
                                                          13
Population Sizing Scalability




Gambler’s ruin model bounds population sizing




                                                14
Run Duration Scalability




Selection-intensity based run-duration model bounds
median convergence time

Drift-time model bounds convergence time

                                                      15
Number of Function Evaluations Scalability




Facetwise models are applicable to rADPs
Testing on adversarial problems bounds performance
of GAs on rADPs

                                                     16
Easy and Hard Scalable Problem Instances

    Easy Scalable               Hard Scalable
      Instances                   Instances




 Easy instances have large signal difference
 Hard instances have very small signal difference



                                                    17
Summary and Conclusions
Empirically analyzed behavior of selectorecombinative GA
with ideal crossover:
   Class of random additively decomposable problems
   Sub-solution fitness sampled from uniform distribution

Verified applicability of facetwise models:
   Developed based on adversarial problems
   GA scales subquadratically with problem size

Analyzed easy and hard problem instances:
   Easy problem instances have large signal, small variance.
   Hard problem instances have small signal, large variance



                                                               18

More Related Content

Viewers also liked

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
 
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
 
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
 

Viewers also liked (16)

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
 
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
 
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
 
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
 
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...
 
iBOA: The Incremental Bayesian Optimization Algorithm
iBOA: The Incremental Bayesian Optimization AlgorithmiBOA: The Incremental Bayesian Optimization Algorithm
iBOA: The Incremental Bayesian Optimization Algorithm
 
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
 
Fitness inheritance in the Bayesian optimization algorithm
Fitness inheritance in the Bayesian optimization algorithmFitness inheritance in the Bayesian optimization algorithm
Fitness inheritance in the Bayesian optimization algorithm
 
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...
 
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
 
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
 
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
 

Similar to Empirical Analysis of ideal recombination on random decomposable problems

Bayesian Multi-topic Microarray Analysis with Hyperparameter Reestimation
Bayesian Multi-topic Microarray Analysis with Hyperparameter ReestimationBayesian Multi-topic Microarray Analysis with Hyperparameter Reestimation
Bayesian Multi-topic Microarray Analysis with Hyperparameter Reestimation
Tomonari Masada
 
MDPSO_SDM_2012_Souma
MDPSO_SDM_2012_SoumaMDPSO_SDM_2012_Souma
MDPSO_SDM_2012_Souma
MDO_Lab
 
DS-004-Robust Design
DS-004-Robust DesignDS-004-Robust Design
DS-004-Robust Design
handbook
 
Mm presentation bkk
Mm presentation bkkMm presentation bkk
Mm presentation bkk
Shannon Chen
 
An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...
An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...
An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...
Shubhashis Shil
 

Similar to Empirical Analysis of ideal recombination on random decomposable problems (20)

Let's get ready to rumble redux: Crossover versus mutation head to head on ex...
Let's get ready to rumble redux: Crossover versus mutation head to head on ex...Let's get ready to rumble redux: Crossover versus mutation head to head on ex...
Let's get ready to rumble redux: Crossover versus mutation head to head on ex...
 
A summary of Categorical Reparameterization with Gumbel-Softmax by Jang et al...
A summary of Categorical Reparameterization with Gumbel-Softmax by Jang et al...A summary of Categorical Reparameterization with Gumbel-Softmax by Jang et al...
A summary of Categorical Reparameterization with Gumbel-Softmax by Jang et al...
 
The Robust Optimization of Non-Linear Requirements Models
The Robust Optimization of Non-Linear Requirements ModelsThe Robust Optimization of Non-Linear Requirements Models
The Robust Optimization of Non-Linear Requirements Models
 
Md Mushfiqul Alam: Biological, NeuralNet Approaches to Recognition, Gain Cont...
Md Mushfiqul Alam: Biological, NeuralNet Approaches to Recognition, Gain Cont...Md Mushfiqul Alam: Biological, NeuralNet Approaches to Recognition, Gain Cont...
Md Mushfiqul Alam: Biological, NeuralNet Approaches to Recognition, Gain Cont...
 
Bayesian Multi-topic Microarray Analysis with Hyperparameter Reestimation
Bayesian Multi-topic Microarray Analysis with Hyperparameter ReestimationBayesian Multi-topic Microarray Analysis with Hyperparameter Reestimation
Bayesian Multi-topic Microarray Analysis with Hyperparameter Reestimation
 
Genetic Algorithms and Genetic Programming for Multiscale Modeling
Genetic Algorithms and Genetic Programming for Multiscale ModelingGenetic Algorithms and Genetic Programming for Multiscale Modeling
Genetic Algorithms and Genetic Programming for Multiscale Modeling
 
Linkage Learning for Pittsburgh LCS: Making Problems Tractable
Linkage Learning for Pittsburgh LCS: Making Problems TractableLinkage Learning for Pittsburgh LCS: Making Problems Tractable
Linkage Learning for Pittsburgh LCS: Making Problems Tractable
 
MDPSO_SDM_2012_Souma
MDPSO_SDM_2012_SoumaMDPSO_SDM_2012_Souma
MDPSO_SDM_2012_Souma
 
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...
 
Cz24655657
Cz24655657Cz24655657
Cz24655657
 
DS-004-Robust Design
DS-004-Robust DesignDS-004-Robust Design
DS-004-Robust Design
 
11.1. PPT on How to crack ML Competitions all steps explained.pptx
11.1. PPT on How to crack ML Competitions all steps explained.pptx11.1. PPT on How to crack ML Competitions all steps explained.pptx
11.1. PPT on How to crack ML Competitions all steps explained.pptx
 
Mm presentation bkk
Mm presentation bkkMm presentation bkk
Mm presentation bkk
 
Deep Advances in Generative Modeling
Deep Advances in Generative ModelingDeep Advances in Generative Modeling
Deep Advances in Generative Modeling
 
Regression vs Deep Neural net vs SVM
Regression vs Deep Neural net vs SVMRegression vs Deep Neural net vs SVM
Regression vs Deep Neural net vs SVM
 
35 38
35 3835 38
35 38
 
An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...
An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...
An Approach to Solve Winner Determination in Combinatorial Reverse Auctions U...
 
Maximizing the Representation Gap between In-domain & OOD examples
Maximizing the Representation Gap between In-domain & OOD examplesMaximizing the Representation Gap between In-domain & OOD examples
Maximizing the Representation Gap between In-domain & OOD examples
 
Do not Match, Inherit: Fitness Surrogates for Genetics-Based Machine Learning...
Do not Match, Inherit: Fitness Surrogates for Genetics-Based Machine Learning...Do not Match, Inherit: Fitness Surrogates for Genetics-Based Machine Learning...
Do not Match, Inherit: Fitness Surrogates for Genetics-Based Machine Learning...
 
Memory Polynomial Based Adaptive Digital Predistorter
Memory Polynomial Based Adaptive Digital PredistorterMemory Polynomial Based Adaptive Digital Predistorter
Memory Polynomial Based Adaptive Digital Predistorter
 

More from kknsastry

Automated alphabet reduction with evolutionary algorithms for protein structu...
Automated alphabet reduction with evolutionary algorithms for protein structu...Automated alphabet reduction with evolutionary algorithms for protein structu...
Automated alphabet reduction with evolutionary algorithms for protein structu...
kknsastry
 
Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...
Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...
Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...
kknsastry
 

More from kknsastry (20)

Population sizing for entropy-based model buliding In genetic algorithms
Population sizing for entropy-based model buliding In genetic algorithms Population sizing for entropy-based model buliding In genetic algorithms
Population sizing for entropy-based model buliding In genetic algorithms
 
Modeling selection pressure in XCS for proportionate and tournament selection
Modeling selection pressure in XCS for proportionate and tournament selectionModeling selection pressure in XCS for proportionate and tournament selection
Modeling selection pressure in XCS for proportionate and tournament selection
 
Modeling XCS in class imbalances: Population sizing and parameter settings
Modeling XCS in class imbalances: Population sizing and parameter settingsModeling XCS in class imbalances: Population sizing and parameter settings
Modeling XCS in class imbalances: Population sizing and parameter settings
 
Substructrual surrogates for learning decomposable classification problems: i...
Substructrual surrogates for learning decomposable classification problems: i...Substructrual surrogates for learning decomposable classification problems: i...
Substructrual surrogates for learning decomposable classification problems: i...
 
Let's get ready to rumble redux: Crossover versus mutation head to head on ex...
Let's get ready to rumble redux: Crossover versus mutation head to head on ex...Let's get ready to rumble redux: Crossover versus mutation head to head on ex...
Let's get ready to rumble redux: Crossover versus mutation head to head on ex...
 
Automated alphabet reduction with evolutionary algorithms for protein structu...
Automated alphabet reduction with evolutionary algorithms for protein structu...Automated alphabet reduction with evolutionary algorithms for protein structu...
Automated alphabet reduction with evolutionary algorithms for protein structu...
 
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
 
Modeling selection pressure in XCS for proportionate and tournament selection
Modeling selection pressure in XCS for proportionate and tournament selectionModeling selection pressure in XCS for proportionate and tournament selection
Modeling selection pressure in XCS for proportionate and tournament selection
 
Modeling XCS in class imbalances: Population size and parameter settings
Modeling XCS in class imbalances: Population size and parameter settingsModeling XCS in class imbalances: Population size and parameter settings
Modeling XCS in class imbalances: Population size and parameter settings
 
Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...
Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...
Fast and accurate reaction dynamics via multiobjective genetic algorithm opti...
 
On Extended Compact Genetic Algorithm
On Extended Compact Genetic AlgorithmOn Extended Compact Genetic Algorithm
On Extended Compact Genetic Algorithm
 
Silicon Cluster Optimization Using Extended Compact Genetic Algorithm
Silicon Cluster Optimization Using Extended Compact Genetic AlgorithmSilicon Cluster Optimization Using Extended Compact Genetic Algorithm
Silicon Cluster Optimization Using Extended Compact Genetic Algorithm
 
A Practical Schema Theorem for Genetic Algorithm Design and Tuning
A Practical Schema Theorem for Genetic Algorithm Design and TuningA Practical Schema Theorem for Genetic Algorithm Design and Tuning
A Practical Schema Theorem for Genetic Algorithm Design and Tuning
 
On the Supply of Building Blocks
On the Supply of Building BlocksOn the Supply of Building Blocks
On the Supply of Building Blocks
 
Don't Evaluate, Inherit
Don't Evaluate, InheritDon't Evaluate, Inherit
Don't Evaluate, Inherit
 
Efficient Cluster Optimization Using A Hybrid Extended Compact Genetic Algori...
Efficient Cluster Optimization Using A Hybrid Extended Compact Genetic Algori...Efficient Cluster Optimization Using A Hybrid Extended Compact Genetic Algori...
Efficient Cluster Optimization Using A Hybrid Extended Compact Genetic Algori...
 
Modeling Tournament Selection with Replacement Using Apparent Added Noise
Modeling Tournament Selection with Replacement Using Apparent Added NoiseModeling Tournament Selection with Replacement Using Apparent Added Noise
Modeling Tournament Selection with Replacement Using Apparent Added Noise
 
Evaluation Relaxation as an Efficiency-Enhancement Technique: Handling Varian...
Evaluation Relaxation as an Efficiency-Enhancement Technique: Handling Varian...Evaluation Relaxation as an Efficiency-Enhancement Technique: Handling Varian...
Evaluation Relaxation as an Efficiency-Enhancement Technique: Handling Varian...
 
Analysis of Mixing in Genetic Algorithms: A Survey
Analysis of Mixing in Genetic Algorithms: A SurveyAnalysis of Mixing in Genetic Algorithms: A Survey
Analysis of Mixing in Genetic Algorithms: A Survey
 
How Well Does A Single-Point Crossover Mix Building Blocks with Tight Linkage?
How Well Does A Single-Point Crossover Mix Building Blocks with Tight Linkage?How Well Does A Single-Point Crossover Mix Building Blocks with Tight Linkage?
How Well Does A Single-Point Crossover Mix Building Blocks with Tight Linkage?
 

Recently uploaded

abortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammam
abortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammamabortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammam
abortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammam
samsungultra782445
 
FOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdf
FOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdfFOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdf
FOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdf
Cocity Enterprises
 
Abortion pills in Saudi Arabia (+919707899604)cytotec pills in dammam
Abortion pills in Saudi Arabia (+919707899604)cytotec pills in dammamAbortion pills in Saudi Arabia (+919707899604)cytotec pills in dammam
Abortion pills in Saudi Arabia (+919707899604)cytotec pills in dammam
samsungultra782445
 
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdfMASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
Cocity Enterprises
 

Recently uploaded (20)

Webinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech BelgiumWebinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech Belgium
 
FE Credit and SMBC Acquisition Case Studies
FE Credit and SMBC Acquisition Case StudiesFE Credit and SMBC Acquisition Case Studies
FE Credit and SMBC Acquisition Case Studies
 
20240419-SMC-submission-Annual-Superannuation-Performance-Test-–-design-optio...
20240419-SMC-submission-Annual-Superannuation-Performance-Test-–-design-optio...20240419-SMC-submission-Annual-Superannuation-Performance-Test-–-design-optio...
20240419-SMC-submission-Annual-Superannuation-Performance-Test-–-design-optio...
 
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
 
abortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammam
abortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammamabortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammam
abortion pills in Riyadh Saudi Arabia (+919707899604)cytotec pills in dammam
 
Technology industry / Finnish economic outlook
Technology industry / Finnish economic outlookTechnology industry / Finnish economic outlook
Technology industry / Finnish economic outlook
 
Test bank for advanced assessment interpreting findings and formulating diffe...
Test bank for advanced assessment interpreting findings and formulating diffe...Test bank for advanced assessment interpreting findings and formulating diffe...
Test bank for advanced assessment interpreting findings and formulating diffe...
 
FOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdf
FOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdfFOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdf
FOREX FUNDAMENTALS: A BEGINNER'S GUIDE.pdf
 
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
 
Dubai Call Girls Deira O525547819 Dubai Call Girls Bur Dubai Multiple
Dubai Call Girls Deira O525547819 Dubai Call Girls Bur Dubai MultipleDubai Call Girls Deira O525547819 Dubai Call Girls Bur Dubai Multiple
Dubai Call Girls Deira O525547819 Dubai Call Girls Bur Dubai Multiple
 
In Sharjah ௵(+971)558539980 *_௵abortion pills now available.
In Sharjah ௵(+971)558539980 *_௵abortion pills now available.In Sharjah ௵(+971)558539980 *_௵abortion pills now available.
In Sharjah ௵(+971)558539980 *_௵abortion pills now available.
 
W.D. Gann Theory Complete Information.pdf
W.D. Gann Theory Complete Information.pdfW.D. Gann Theory Complete Information.pdf
W.D. Gann Theory Complete Information.pdf
 
Shrambal_Distributors_Newsletter_May-2024.pdf
Shrambal_Distributors_Newsletter_May-2024.pdfShrambal_Distributors_Newsletter_May-2024.pdf
Shrambal_Distributors_Newsletter_May-2024.pdf
 
Abortion pills in Saudi Arabia (+919707899604)cytotec pills in dammam
Abortion pills in Saudi Arabia (+919707899604)cytotec pills in dammamAbortion pills in Saudi Arabia (+919707899604)cytotec pills in dammam
Abortion pills in Saudi Arabia (+919707899604)cytotec pills in dammam
 
NO1 Verified Online Love Vashikaran Specialist Kala Jadu Expert Specialist In...
NO1 Verified Online Love Vashikaran Specialist Kala Jadu Expert Specialist In...NO1 Verified Online Love Vashikaran Specialist Kala Jadu Expert Specialist In...
NO1 Verified Online Love Vashikaran Specialist Kala Jadu Expert Specialist In...
 
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize ThemSignificant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
 
falcon-invoice-discounting-unlocking-prime-investment-opportunities
falcon-invoice-discounting-unlocking-prime-investment-opportunitiesfalcon-invoice-discounting-unlocking-prime-investment-opportunities
falcon-invoice-discounting-unlocking-prime-investment-opportunities
 
fundamentals of corporate finance 11th canadian edition test bank.docx
fundamentals of corporate finance 11th canadian edition test bank.docxfundamentals of corporate finance 11th canadian edition test bank.docx
fundamentals of corporate finance 11th canadian edition test bank.docx
 
Lion One Corporate Presentation May 2024
Lion One Corporate Presentation May 2024Lion One Corporate Presentation May 2024
Lion One Corporate Presentation May 2024
 
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdfMASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
 

Empirical Analysis of ideal recombination on random decomposable problems

  • 1. Empirical Analysis of Ideal Crossover on Random Additively Decomposable Problems Kumara Sastry1, Martin Pelikan2, David E. Goldberg1 1Illinois Genetic Algorithms Laboratory (IlliGAL) University of Illinois at Urbana-Champaign, Urbana, IL 61801 2MissouriEstimation of Distribution Algorithm Lab (MEDAL) University of Missouri at St. Louis, St. Louis, MO ksastry@uiuc.edu, pelikan@cs.umsl.edu, deg@uiuc.edu http://www.illigal.uiuc.edu, http://medal.cs.umsl.edu Supported by AFOSR FA9550-06-1-0096, NSF DMR 03-25939, and CAREER ECS-0547013. Computational results obtained using CSE’s Turing cluster.
  • 2. Roadmap Adversarial test problem design Random additively decomposable problems Ideal crossover Scalability of selectorecombinative GAs Population sizing and Run duration Experimental Procedure Key Results Summary and Conclusions 2
  • 3. Adversarial Test Problem Design Test systems on boundary of design envelope Common approach in designing complex systems GAs are complex systems [Goldberg, 2002] GA design envelope characterized by different dimensions of problem difficulty Thwart the mechanism of GAs to the extreme Fluctuating R P Noise Deception Scaling 3
  • 4. Random Additively Decomposable Problem Focus on nearly decomposable problems [Simon, 1960] Three desired features Scalability: Able to control problem size and difficulty Known optimum: Allows comparison of different solvers Easy problem instance generation rADP fitness function: Si represents variable subset for ith subproblem Each subset consists of k bits gi is the fitness of the ith subproblem gi is sampled from uniform distribution U[0,1] 4
  • 5. Ideal Crossover: Exchange Building Blocks Population sizing and run duration models assume good exchange of building blocks Simulate what we ideally want to achieve with model- building GAs For example, extended compact GA [Harik, 1999] Ideal recombination operator Effectively exchange building blocks Don’t disrupt any building block Uniform building-block-wise crossover BBs #1 and #3 exchanged Exchange BBs with probability 0.5 5
  • 6. Purpose: Analyze Ideal Crossover on rADPs Analyze behavior of selectorecombinative GAs on rADPs Verify the validity of lessons learned from adversarial test problems Expand the pool of test problems 6
  • 7. Selectorecombinative GA Population Sizing Noise-to-fitness variance ratio Error tolerance # Components (# BBs) Signal-to-Noise ratio # Competing sub-components Gambler’s ruin model [Harik, et al, 1997] Combines decision making and supply models Additive Gaussian noise with variance σ2N 7
  • 8. GA Run Duration (Selection) Selection-Intensity based model [Bulmer, 1980; Mühlenbein & Schlierkamp-Voosen, 1993; Thierens & Goldberg, 1994; Bäck, 1994; Miller & Goldberg, 1995 & 1996] Derived for the OneMax problem Applicable to additively-separable problems [Miller, 1997] Problem size (m·k ) Selection Intensity [Miller & Goldberg, 1995; Sastry & Goldberg, 2002] 8
  • 9. GA Run Duration (Drift) Accumulation of stochastic errors due to finite population Proportion of competing sub-solutions change due to drift Drift time [Goldberg & Segrest, 1987]: Substituting population sizing bound 9
  • 10. Signal-to-Noise Ratio for rADPs Signal d is the fitness difference between best and second best sub-solutions jth order statistic follows a Beta distribution with α = j and β = 2k-j+1 Probability density function (p.d.f) of d: p.d.f. of sub-solution fitness variance approximation E[1/d] = 2k and E[σ2BB] ≈ 1/12 10
  • 11. Assumptions and Experimental Setup Non-overlapping sub-problems Identical sub-problems across different partitions g1 = g2 = … = gm Selectorecombinative GA Binary tournament selection 10,000 random problem instances m = 5 – 50, k = 3, 4, and 5 Minimum population size determined by bisection method Population correctly converges to at least m-1 out of m BBs in 49 out of 50 independent runs Averaged over 30 bisection runs Results averaged over 1,500 GA runs 11
  • 12. Population Sizing & Run Duration Histograms Population size Run duration m = 50 m = 50 Tail increases with m 0.15-0.59% of rADP instances require # evals greater than 3σ from the median 12
  • 13. Easy and Hard Problem Instances Hard instance Subsolution fitness Min signal Max noise Sorted subsolution index Easy instance Subsolution fitness Max signal Min noise Sorted subsolution index 13
  • 14. Population Sizing Scalability Gambler’s ruin model bounds population sizing 14
  • 15. Run Duration Scalability Selection-intensity based run-duration model bounds median convergence time Drift-time model bounds convergence time 15
  • 16. Number of Function Evaluations Scalability Facetwise models are applicable to rADPs Testing on adversarial problems bounds performance of GAs on rADPs 16
  • 17. Easy and Hard Scalable Problem Instances Easy Scalable Hard Scalable Instances Instances Easy instances have large signal difference Hard instances have very small signal difference 17
  • 18. Summary and Conclusions Empirically analyzed behavior of selectorecombinative GA with ideal crossover: Class of random additively decomposable problems Sub-solution fitness sampled from uniform distribution Verified applicability of facetwise models: Developed based on adversarial problems GA scales subquadratically with problem size Analyzed easy and hard problem instances: Easy problem instances have large signal, small variance. Hard problem instances have small signal, large variance 18