liacs natural computing group leiden university evolutionary algorithms

LIACS Natural Computing Group Leiden University

Evolutionary Algorithms

LIACS Natural Computing Group Leiden University2

Overview

• Introduction: Optimization and EAs• Genetic Algorithms• Evolution Strategies• Theory• Examples

Background IBiology = Engineering (Daniel Dennett)

Introduction • Modeling

• Simulation

• Optimization

! ! ! ! ! !???

! ! ! ???! ! !

??? ! ! !! ! !

Input: Known (measured)Output: Known (measured)

Interrelation: Unknown

Input: Will be given

Model: Already exists

How is the result for the input?

Objective: Will be givenHow (with which parameter settings)to achieve this objective?

Simulation vs. Optimization

SimulatorSimulator

… what happens if?

Result

Trial & Error

SimulatorSimulator

... how do I achieve the best result?

OptimalResult

Maximization / MinimizationIf so, multiple objectives

OptimizerOptimizer

Introduction:

OptimizationEvolutionary Algorithms

Optimization

• f : objective function

– High-dimensional

– Non-linear, multimodal

– Discontinuous, noisy, dynamic

• M M1 M2 ... Mn heterogeneous

• Restrictions possible over

• Good local, robust optimum desired

• Realistic landscapes are like that!Global Minimum

Local, robust optimum

• Illustrative Example: Optimize Efficiency– Initial:

– Evolution:

• 32% Improvement in Efficiency !

Optimization Creating Innovation

Dynamic Optimization

• Dynamic Function• 30-dimensional• 3D-Projection

Classification of Optimization Algorithms• Direct optimization algorithm:

Evolutionary Algorithms

• First order optimization algorithm:

e.g., gradient method

• Second order optimization algorithm:

e.g., Newton method

Iterative Optimization Methods

General description:Actual Point

New Point

Directional vector

Step size (scalar)

3x• At every Iteration:

– Choose direction

– Determine step size

• Direction:

– Gradient

– Random

• Step size:

– 1-dim. optimization

– Random

– Self-adaptive

• Global convergence with probability one:

• General, but for practical purposes useless

• Convergence velocity:

• Local analysis only, specific (convex) functions

1))(*Pr(lim

)))(())1((( maxmax tPftPfE

The Fundamental Challenge

• Global convergence (with probability 1):

• General statement (holds for all functions)

• Useless for practical situations:

– Time plays a major role in practice

– Not all objective functions are relevant in practice

1))(*Pr(lim

Theoretical Statements

f(x1,x2)

(x*1,x*2)

f(x*1,x*2)

An Infinite Number of Pathological Cases !

• NFL-Theorem:

– All optimization algorithms perform equally well iff performance is averaged over all possible optimization problems.

• Fortunately: We are not Interested in „all possible problems“

• Very specific statements

– Convex objective functions

– Describes convergence in local optima

– Very extensive analysis for Evolution Strategies

)))(())1((( maxmax tPftPfE

Theoretical Statements

Evolutionary AlgorithmPrinciples

Model-Optimization-Action

Evaluation

Optimizer

Business Process Model

Simulation

i scale

desiredcalculatedweightquality

Function Model from Data

Experiment SubjectiveFunction(s)

...)( yfi

Evolutionary Algorithms TaxonomyEvolutionary AlgorithmsEvolutionary Algorithms

Evolution StrategiesEvolution Strategies Genetic AlgorithmsGenetic Algorithms OtherOther

• Evolutionary Progr.• Differential Evol.• GP• PSO• EDA• Real-coded Gas• …

• Mixed-integer capabilities

• Emphasis on mutation

• Self-adaptation

• Small population sizes

• Deterministic selection

• Developed in Germany

• Theory focused on convergence velocity

• Discrete representations

• Emphasis on crossover

• Constant parameters

• Larger population sizes

• Probabilistic selection

• Developed in USA

• Theory focused on schema processing

Generalized Evolutionary Algorithm

t := 0;

initialize(P(t));

evaluate(P(t));

while not terminate do

P‘(t) := mating_selection(P(t));

P‘‘(t) := variation(P‘(t));

evaluate(P‘‘(t));

P(t+1) := environmental_selection(P‘‘(t) Q);

t := t+1;

Genetic Algorithms

Genetic Algorithms: Mutation

• Mutation by bit inversion with probability pm.

• pm identical for all bits.

• pm small (e.g., pm = 1/n).

0 1 1 1 0 1 0 1 0 0 0 0 0 01

0 1 1 1 0 0 0 1 0 1 0 0 0 01

Genetic Algorithms: Crossover

• Crossover applied with probability pc.

• pc identical for all individuals.

• k-point crossover: k points chosen randomly.• Example: 2-point crossover.

Genetic Algorithms: Selection

• Fitness proportional:– f fitness, population size

• Tournament selection:– Randomly select q << individuals.– Copy best of these q into next generation.– Repeat times.– q is the tournament size (often: q = 2).

Some Theory

jfljfl

kafamfkp

))())((Pr()(

Convergence Velocity Analysis• (1+1)-GA, (1,)-GA, (1+)-GA• For counting ones function• Convergence velocity:

)1)((2

iji ntE )(

1)()( QInN ij

• Optimum mutation rate ?

• Absorption times from transition matrix

in block form, using

Convergence Velocity Analysis II

Convergence Velocity Analysis III• P too large:

– Exponential• P too small:

– Almost constant• Optimal:

– O(l ln l)

• (1,l)-GA (kmin = -fa), (1+l)-GA (kmin = 0) :

Convergence Velocity Analysis IV

Convergence Velocity Analysis V• (1,)-GA, (1GA (1,)-ES, (1ES

• Conclusion: Unifying, search-space independent theory ?

kpkafl

Convergence Velocity Analysis VI• (,)-GA, (GA

• Theory• Experiment

Evolution Strategies

Evolution Strategy – Basics

• Mostly real-valued search space IRn

– also mixed-integer, discrete spaces• Emphasis on mutation

– n-dimensional normal distribution– expectation zero

• Different recombination operators• Deterministic selection

– ()-selection: Deterioration possible– ()-selection: Only accepts improvements

• , i.e.: Creation of offspring surplus• Self-adaptation of strategy parameters.

Representation of search points

)),,...,(( 1 nxxa

),...,( 1 nxxa

• Simple ES with 1/5 success rule:

– Exogenous adaptation of step size

– Mutation: N(0, )

• Self-adaptive ES with single step size:

– One controls mutation for all xi

– Mutation: N(0, )

Representation of search points

)),...,(),,...,(( 11 nnxxa

)),...,(),,...,(),,...,(( 2/)1(111 nnnnxxa

• Self-adaptive ES with individual step sizes:

– One individual i per xi

– Mutation: Ni(0, i)

• Self-adaptive ES with correlated mutation:

– Individual step sizes

– One correlation angle per coordinate pair

– Mutation according to covariance matrix: N(0, C)

Evolution Strategy:

AlgorithmsMutation

Operators: Mutation – one

))1,0(exp(

)),,...,((

• Self-adaptive ES with one step size:

– One controls mutation for all xi

– Mutation: N(0, )Individual before mutation

Individual after mutation

1.: Mutation of step sizes

2.: Mutation of objective variables

Here the new ‘ is used!

• Thereby 0 is the so-called learning rate

– Affects the speed of the -Adaptation

– 0 bigger: faster but more imprecise

– 0 smaller: slower but more precise

– How to choose 0?

– According to recommendation of Schwefel*:

*H.-P. Schwefel: Evolution and Optimum Seeking, Wiley, NY, 1995.

Position of parents (here: 5)

Offspring of parent lies on the hyper sphere (for n > 10);Position is uniformly distributed

Contour lines of objective function

Pros and Cons: One • Advantages:

– Simple adaptation mechanism

– Self-adaptation usually fast and precise

• Disadvantages:

– Bad adaptation in case of complicated contour lines

– Bad adaptation in case of very differently scaled object variables

• -100 < xi < 100 and e.g. -1 < xj < 1

Operators: Mutation – individual i

))1,0()1,0(exp(

)),...,(),,...,((

• Self-adaptive ES with individual step sizes:

– One i per xi

– Mutation: Ni(0, i) Individual before Mutation

Individual after Mutation

1.: Mutation of individual step sizes

2.: Mutation of object variables

The new individual i‘ are used here!

• , ‘ are learning rates, again– ‘: Global learning rate– N(0,1): Only one realisation– : local learning rate

– Ni(0,1): n realisations

– Suggested by Schwefel*:

*H.-P. Schwefel: Evolution and Optimum Seeking, Wiley, NY, 1995.

Position of parents (here: 5)

Offspring are located on the hyperellipsoid (für n > 10);position equally distributed.

Contour lines

Pros and Cons: Individual i

• Advantages:

– Individual scaling of object variables

– Increased global convergence reliability

• Disadvantages:

– Slower convergence due to increased learning effort

– No rotation of coordinate system possible• Required for badly conditioned objective function

Operators: Correlated Mutations

))1,0()1,0(exp(

)),...,(),,...,(),,...,((

2/)1(111

Individual before mutation

Individual after mutation

1.: Mutation of Individual step sizes

2.: Mutation of rotation angles

New convariance matrix C‘ used here!

• Self-adaptive ES with correlated mutations:

– Individual step sizes

– One rotation angle for each pair of coordinates

– Mutation according to covariance matrix: N(0, C)

3.: Mutation of object variables

Operators: Correlated Mutations

)2tan()(2

1 22)( ijjijiijc

• Interpretation of rotation angles ij

• Mapping onto convariances according to

Operators: Correlated Mutation

)(2 jjjj sign

, ‘, are again learning rates ‘ as before = 0,0873 (corresponds to 5 degree) Out of boundary correction:

Correlated Mutations for ES

Position of parents (hier: 5)

Offspring is located on the Rotatable hyperellipsoid(for n > 10); position equallydistributed.

Contour lines

Operators: Correlated Mutations),0( CN

• How to create ?

– Multiplication of uncorrelated mutation vector with n(n-1)/2 rotational matrices

– Generates only feasible (positiv definite) correlation matrices

ijijc R

Operators: Correlated Mutations• Structure of rotation matrix

)cos(1

Operators: Correlated Mutations• Implementation of correlated mutations

nq := n(n-1)/2;for i:=1 to n do

u[i] := [i] * Ni(0,1);

for k:=1 to n-1 don1 := n-k;n2 := n;for i:=1 to k do

d1 := u[n1]; d2:= u[n2];u[n2] := d1*sin([nq])+ d2*cos([nq]);u[n1] := d1*cos([nq])- d2*sin([nq]);n2 := n2-1;nq := nq-1;

Generation of the uncorrelated mutation vector

Rotations

Pros and Cons: Correlated Mutations

• Advantages:

– Individual scaling of object variables

– Rotation of coordinate system possible

– Increased global convergence reliability

• Disadvantages:

– Much slower convergence

– Effort for mutations scales quadratically

– Self-adaptation very inefficient

Operators: Mutation – Addendum

• Generating N(0,1)-distributed random numbers?

)log(2

1)1,0[2

If w > 1

)1,0(~, 21 Nxx

Evolution Strategy:

AlgorithmsRecombination

Operators: Recombination• Only for > 1

• Directly after Secektion

• Iteratively generates offspring:

for i:=1 to dochoose recombinant r1 uniformly at random

from parent_population;choose recombinant r2 <> r1 uniformly at random

from parent population;offspring := recombine(r1,r2);add offspring to offspring_population;

Operators: Recombination• How does recombination work?

• Discrete recombination:

– Variable at position i will be copied at random (uniformly distr.) from parent 1 or parent 2, position i.Parent 1

Parent 2

Offspring

Operators: Recombination• Intermediate recombination:

– Variable at position i is arithmetic mean ofParent 1 and Parent 2, position i.

Parent 1

Parent 2

Offspring

2/)( 1,1, 21 rr xx

Operators: Recombination

Parent 1

Parent 2

Offspring

…Parent

• Global discrete recombination:

– Considers all parents

Operators: Recombination

Parent 1

Parent 2

Offspring

…Parent

• Global intermediary recombination:

– Considers all parents1,2r

Evolution Strategy

AlgorithmsSelection

Operators: ()-Selection

• ()-Selection means:

– parents produce offspring by• (Recombination +)

• Mutation

– These individuals will be considered together

– The best out of will be selected („survive“)• Deterministic selection

– This method guarantees monotony• Deteriorations will never be accepted

= Actual solution candidate= New solution candidate

Recombination may be left outMutation always exists!

Operators: ()-Selection

• ()-Selection means:

– parents produce offspring by• (Recombination +)

• Mutation

– offspring will be considered alone

– The best out of offspring will be selected• Deterministic selection

– The method doesn‘t guarantee monotony• Deteriorations are possible

• The best objective function value in generation t+1 may be worse than the best in generation t.

Operators: Selection• Example: (2,3)-Selection

• Example: (2+3)-Selection

Parents don‘t survive!Parents don‘t survive …

… but a worse offspring.

… now this offspring survives.

Exception!

• Possible occurrences of selection– (1+1)-ES: One parent, one offspring, 1/5-Rule– (1,)-ES: One Parent, offspring

• Example: (1,10)-Strategy• One step size / n self-adaptive step sizes • Mutative step size control• Derandomized strategy

– ()-ES: > 1 parents, offspring• Example: (2,15)-Strategy• Includes recombination• Can overcome local optima

– (+)-strategies: elitist strategies

Operators: Selection

Evolution Strategy:

Self adaptation ofstep sizes

Self-adaptation

• No deterministic step size control!

• Rather: Evolution of step sizes

– Biology: Repair enzymes, mutator-genes

• Why should this work at all?

– Indirect coupling: step sizes – progress

– Good step sizes improve individuals

– Bad ones make them worse

– This yields an indirect step size selection

Self-adaptation: Example

• How can we test this at all?

• Need to know optimal step size …

– Only for very simple, convex objective functions

– Here: Sphere model

– Dynamic sphere model• Optimum locations changes occasionally

)()( i

ii xxxf

: Optimum*x

Self-adaptation: ExampleObjective function value

… and smallest step size measured in the population

average …

Largest …

According to theoryff optimal step sizes

Self-adaptation

• Self-adaptation of one step size

– Perfect adaptation

– Learning time for back adaptation proportional n

– Proofs only for convex functions

• Individual step sizes

– Experiments by Schwefel

• Correlated mutations

– Adaptation much slower

Evolution Strategy:

Derandomization

• Goals:

– Fast convergence speed

– Fast step size adaptation

– Precise step size adaptation

– Compromise convergence velocity – convergence reliability

• Idea: Realizations of N(0, ) are important!

– Step sizes and realizations can be much different from each other

– Accumulates information over time

Derandomzed (1,)-ES

scalggg

N Zxxk

• Current parent: in generation g

• Mutation (k=1,…,): Offspring k

Global step size in generation g

Individual step sizes in generation g

)1,0(~),...,( 1 NzzzZ in

• Selection: Choice of best offspring

selxx 1 Best of offspring

in generation g

Derandomized (1,)-ES

gA ZcZcZ

• Accumulation of selected mutations:

The particular mutation vector,which created the parent!

• Also: weighted history of good mutation vectors!• Initialization:

• Weight factor:00

Derandomized (1,)-ES

Z gAgg

• Step size adaptation: Norm of vector

Vector of absolute values

1 Regulates adaptation

speed and precision

Derandomized (1,)-ES• Explanations:

– Normalization of average variations in case of missing selection (no bias):

– Correction for small n: 1/(5n)

– Learning rates:

scal /1

Evolution Strategy:

Rules of thumb

Some Theory Highlights

• For (1,)-strategies:

• For ()-strategies (discrete and intermediary recombination):

Problem dimensionality

Speedup by is just logarithmic – more processors are only to a limited extend useful to increase

ln~ Genetic Repair Effect

of recombination!

Genetic Repair Effectof recombination!

• For strategies with global intermediary recombination:

• Good heuristic for (1,):

9.5219.03150

9.4118.82140

9.3018.60130

9.1818.36120

9.0518.10110

8.9117.82100

8.7517.5090

8.5717.1580

8.3716.7570

8.1416.2860

7.8715.7450

7.5315.0740

7.1014.2030

6.4912.9920

5.4510.9110

Mixed-IntegerEvolution Strategies

Mixed-Integer Evolution Strategy

• Generalized optimization problem:

Mixed-Integer ES: Mutation

Learning rates (global)

Geometrical distribution

Mutation Probabilities

Some Application Examples

Safety Optimization – Pilot Study

• Aim: Identification of most appropriate Optimization Algorithm for realistic example!

• Optimizations for 3 test cases and 14 algorithms were performed (28 x 10 = 280 shots)

– Body MDO Crash / Statics / Dynamics– MCO B-Pillar– MCO Shape of Engine Mount

• NuTech’s ES performed significantly better than Monte-Carlo-scheme, GA, and Simulated Annealing

• Results confirmed by statistical hypothesis testing

MDO Crash / Statics / Dynamics

• Minimization of body mass • Finite element mesh

– Crash ~ 130.000 elements– NVH ~ 90.000 elements

• Independent parameters: Thickness of each unit: 109

• Constraints: 18

Algorithm Avg. reduction (kg) Max. reduction (kg) Min. reduction (kg)

Best so far -6.6 -8.3 -3.3

NuTech ES -9.0 -13.4 -6.3

MCO B-Pillar – Side Crash

• Minimization of mass & displacement

• Finite element mesh– ~ 300.000 elements

• Independent parameters: Thickness of 10 units

• Constraints: 0

• ES successfully developed Pareto front

MCO Shape of Engine Mount

• Mass minimal shape withaxial load > 90 kN

• Finite element mesh– ~ 5000 elements

• Independent parameters:9 geometry variables

• Dependent parameters: 7• Constraints: 3• ES optimized mount

– less weight than mount optimized with best so far method

– geometrically better deformation

Safety Optimization – Example of use

• Production Run !• Minimization of body mass • Finite element mesh

– Crash ~ 1.000.000 elements– NVH ~ 300.000 elements

• Independent parameters: – Thickness of each unit: 136

• Constraints: 47, resulting from various loading cases• 180 (10 x 18) shots ~ 12 days• No statistical evaluation due to problem complexity

Safety Optimization – Example of use

• 13,5 kg weight reduction by NuTech’s ES• ~ 2 kg more mass reduction than Best so far method• Typically higher convergence velocity of ES

~ 45% less time (~ 3 days saving) for comparable quality needed• Still potential of improvements after 180 shots.• Reduction of development time from 5 to 2 weeks allows for process

integration

Best so far Method

Generations

Initial Value

Generations

Initial Value

NuTech’s Evolution Strategy

MDO – Optimization Run

107,13%

105,05%

100,00%

All Individuals

Number of Individuals

Fitness Infeasible

Feasible

Optimization of metal stamping process

• Objective: Minimization of defects in the produced parts.

• Variables: Geometric parameters and forces.

• ES finds very good results in short time

• Computationally expensive simulation

Bridgeman Casting Process

• Objective: Max. homogeneity of workpiece after Casting Process

• Variables: 18 continuous speed variables for Casting Schedule

• Computationally expensive simulation (up to 32h simulation time)

Initial (DoE) GCM (Commercial Gradient Based Method)

Evolution Strategy

Turbine Bladeafter Casting

Algorithm 100 OFE

200 OFE

1000OFE

SQP 148 67.7 59

Conjugate directions

135 135 135

Pattern search

157 147 48

DSCP/Rosenbrock

88 38 29

Complex strategy

152 152 111

(1,3)-DES 91 67 67

(1,5)-DES 53 48 48

(1,7)-DES 110 58 58

(1,10)-DES 152 80 28

MAES 122 65 12

Minimize thedeviation of temperture atthe cube´s Surface to1000°C !

Input: TemperatureProfile (12 Variables7 Temperatures and 5 Time-Steps)

Steel Cube Temperature Control

- Maximisation of growth speed- Minimisation of diameter differences

FLUENT: Simulation of fluid flowCASTS: Calculation of Temperatureand concentration field

0 m/s 3 m/s1.5 m/s

Reaction Gas TCS Growing High Purity Silicon Rod

Optimsation of 15 process parameters:

Production time 35 hours 30 hours

Siemens C Reactor

Multipoint Airfoil Optimiziation• Objective: Find pressure profiles that are a compromise between two given

target pressure distributions under two given flow conditions!

• Variables: 12 to 18 Bezier points for the airfoil

High Lift!

Low Drag!

Cruise

Traffic Light Control Optimization

• Objective: Minimization of total delay / number of stops• Variables: Green times for next switching schedule• Dynamic optimization, depending on actual traffic • Better results (3-5%)• Higher flexibility than with traditional controllers

Optimization of elevator control

• Minimization of passenger waiting times.• Better results (~10%) than with traditional controller.• Dynamic optimization, depending on actual traffic.

Optimization of network routing

• Minimization of end-to-end-blockings under service constraints.

• Optimization of routing tables for existing, hard-wired networks.

• 10%-1000% improvement.

Automatic battery configuration

• Configuration and optimization of industry batteries.

• Based on user specifications, given to the system.

• Internet-Configurator (Baan, Hawker, NuTech).

Optimization of reactor fueling.

• Minimization of total costs.

• Creates new fuel assembly reload patterns.

• Clear improvements (1%-5%)of existing expert solutions.

• Huge cost saving.

Optical Coatings: Design Optimization

• Nonlinear mixed-integer problem, variable dimensionality.

• Minimize deviation from desired reflection behaviour.

• Excellent synthesis method; robust and reliable results.

• MOC-Problems: anti-reflection coating, dialetric filters

• Robust design: Sharp peaks vs. Robust peaks

Example: IntravascularUltrasound Image Analysis

• Real-time high-resolution tomographic images from the inside of coronary vessels:

Intravascular Ultrasound Image Analysis

• Detected Features in an IVUS Image:

Vessel Border

Sidebranch

Shadow

Plague

Experimental Results on IVUS images• Parameters for the lumen feature detector

Intravascular Ultrasound Image Analysis: Results

• On each of 5 data sets algorithm ran for 2804 evaluations – 19,5h of total computing time

• Significant improvement over expert tuning

Performance of the best found MI-ES parameter solutions

• A paired two-tailed t-test was performed on the difference measurements for each image dataset using a 95% confidence interval (p=0.05)

• The null-hypothesis is that the mean difference results of the best MI-ES individual and the default parameters are equal.

Literature

• H.-P. Schwefel: Evolution and Optimum Seeking, Wiley, NY, 1995.

• I. Rechenberg: Evolutionsstrategie 94, frommann-holzboog, Stuttgart, 1994.

• Th. Bäck: Evolutionary Algorithms in Theory and Practice, Oxford University Press, NY, 1996.

• Th. Bäck, D.B. Fogel, Z. Michalewicz (Hrsg.): Handbook of Evolutionary Computation, Vols. 1,2, Institute of Physics Publishing, 2000.

liacs natural computing group leiden university evolutionary algorithms

optimization simulator

possible optimization

probability pc

local analysis

pc identical

robust optimumoptimization

practical situations

general statement

Documents

liacs summer 2003 newsletter

playing games - liacs

leiden institute of advanced computer science master study...

universiteit leiden opleiding informatica - liacs

speech recognition liacs media lab leiden university seminar...

universiteit leiden ict in business · 2014-10-31 · 2nd...

affective computing and human robot interaction a short...

self-adaptive vision agents using evolutionary...

aspect-oriented programming in php - leiden...

the mir flickr retrieval evaluation - leiden university ·...

modular design and the importance of models of computation...

leiden university it in business - liacs thesis...

leiden university. the university to discover. enhancing...

seminar speech recognition project support e.m. bakker...

leiden university ict in business - liacs thesis repository

seminarium on component-based software engineering feraaz...

software architecture documentation - liacs

data mining in bioinformatics erwin m. bakker liacs leiden...

blackboard systems - liacs

blackboard systems in socca - introductie - liacs - leiden