Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

107
Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search

Transcript of Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Page 1: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Frédéric Saubion LERIA

Learning and Intelligent OptimizatioN Conference

Autonomous Search

Page 2: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Based on joint works on this topic with :

G. Di TolloA. Fiahlo Y. Hamadi F. Lardeux J. Maturana E. MonfroyM. SchoenauerM. Sebag

Learning and Intelligent OptimizatioN Conference

Page 3: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

1. Introduction2. Main Ideas3. Taxonomy of AS4. Focus on examples5. Conclusion and challenges

Outline

Page 4: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionGeneric modeling tools for engineers

(Decision) VariablesDomains Constraints Mathematical

Model

Solver

Solving Constraint Optimization and Satifaction Problems

Page 5: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionMap coloring problem

Satifaction Problems

Page 6: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionMap coloring problem

Satifaction Problems

Page 7: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionMap coloring problem

Satifaction Problems

Page 8: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Optimization Problems

Travelling Salesman Problem : finda round trip across cities with minimal cost

Page 9: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Optimization Problems

Travelling Salesman Problem : finda round trip across cities with minimal cost

Page 10: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Optimization Problems

Travelling Salesman Problem : finda round trip across cities with minimal cost

Page 11: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionSearch landscapes are difficult to explore

Many variablesComplex constraints

Problems are more and more complex

Page 12: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionSearch landscapes are difficult to explore

Exploration vs. Exploitation Balance

Problems are more and more complex

Page 13: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Boolean Variable {0,1}

IntroductionAn illustrative example : solving SAT

SAT CNF instance

Devising more and more complex Solving algorithms

ba

ca

cba

Litterals

Clauses

Assignment

(1 0 0)

Page 14: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

How to explore the binary search space (assignments) to find a solution ?

Page 15: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Use Local Search

Page 16: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Basic Local Search

0 1 0 1 1

Choose a random initial assignment

Page 17: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Basic Local Search

Compute the number of true and false clauses

Page 18: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Basic Local Search

Try to improve by changing a value (flip)

0 1 0 1 1

0 1 1 1 1

Move to a neighbor

Page 19: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Basic Local Search

Until finding a solution

Page 20: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Short overview of the story : a first greedy version GSATBart Selman, Hector J. Levesque, David G. Mitchell: A New Method for Solving Hard Satisfiability Problems.AAAI 1992: 440-446

A first boat for binary seas

Page 21: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

10 1 0 1 1 0 0

1 1 1 1

0 0 0

Problem : Many possible moves (many variables)

Page 22: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Restrict neighborhoodSelect a false clause C

fda a b c d e f g

0 1 0 1 1 0 0

Page 23: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Get stuck in local optima

Page 24: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Add pertubationsSelect a false clause C

With a random probability p Perform a random flip for C

With (1-p) Select the variable with best IMP Perform best move

If solution then stopElse go on

Parameter !

Page 25: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Use restarts

False Clauses

Iterations

Parameter !

Page 26: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

WalkSAT : adding a noise and random restart

Henry A. Kautz, Bart Selman: Noise Strategies for Improving Local Search..AAAI 1994

Page 27: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

How to break ties ?

0 1 0 1 1 0 0

+3 +3

+3

Page 28: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Add more sophisticated heuristics Compute the age of the variableIf the best variable is not the most recent then flipElseWith a random probability p’

Perform a random flip the second best

With (1-p’) Flip the best

Parameter !

Page 29: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Novelty : using more strategies to perform improvements (age of the variable)D.A. McAllester, B. Selman and H. Kautz. Evidence for invariant in local search.In Proceedings of AAAI-97, AAAI Press 1997, pages 321-326.

Page 30: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

And improvements go on …

Novelty +,Novelty ++, …, TNM, Sattime…

Page 31: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Introduction

Devising more and more complex Solving algorithms

Captain Jack : many indicators and thus selection strategies

Dave A. D. Tompkins, Adrian Balint, Holger H. Hoos: Captain Jack: New Variable Selection Heuristics in Local Search for SAT. SAT 2011: 302-316

Page 32: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

IntroductionAdding more parameters and heuristics

Devising more and more complex Solving algorithms

More flexible algorithms Fit to different instances

Set parameters/heuristics values Understand the behavior

Page 33: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

John Rice. The algorithm selection problem. Technical Report CSD-TR152, Computer science department, Purdue University, 1975.

The Algorithm Selection Problem

Main ideas

Page 34: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

John Rice. The algorithm selection problem. Technical Report CSD-TR152, Computer science department, Purdue University, 1975.

The Algorithm Selection Problem

Main ideas

Page 35: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Related Questions

Main ideas

Using several algorithms for solving a classof problems

Page 36: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Related Questions

Main ideas

Adjusting the parametersof one algorithm

Page 37: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Main Objectives

Main ideas

Need for more autonomous solving tools

Increasing number of works in this trend : LION, Specialsessions in EA conferences (GECCO,…) …

Page 38: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Ideas for More Autonomous Solvers

How to use an algorithm that includes

•Many parameters

•Many possible heuristics or components

Ideas

Page 39: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Ideas for More Autonomous Solvers

How to use an algorithm that include

•Many parameters

•Many possible heuristics or components

How to automate all these choices ?

Ideas

Page 40: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Off-line Automated Tuning

Ideas

Run your solver on some problems

Experiment automatically parameters values

Page 41: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Off-line Automated Tuning

Ideas

Run your solver on new problems with these parameters values

Page 42: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Off-line Automated Tuning

Ideas

Question : Generality of the parameters ?

Page 43: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

On-line Parameter Control

Ideas

Try to react during the resolution by changing the parameter

Page 44: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

On-line Parameter Control

Ideas

Example : try to increase some parameter when possible

Page 45: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

On-line Parameter Control

Ideas

Question : How to react efficiently ?

Page 46: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Hyper Heuristics

Ideas

Combine basic solving heuristics

Page 47: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Hyper Heuristics

Ideas

Get new solvers

Page 48: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Hyper Heuristics

Ideas

Question : How to learn the suitable solver ?

Page 49: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Portfolios Based Solvers

Ideas

Use different types of solvers

Page 50: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Portfolios Based Solvers

Ideas

Learn how to select the right solver for a given problem

Page 51: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Portfolios Based Solvers

Ideas

Question : Reliability of the learning process ?

Page 52: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Evolutionary ComputationA. E. Eiben, R. Hinterding, and Z. Michalewicz. Parameter control in

evolutionary algorithms. IEEE Trans. Evolutionary Computation,

3(2) :124141, 1999.

Reactive Search Battiti R, Brunato M, Mascia F (2008) Reactive Search and Intelligent

Optimization, ORCS interf., vol 45. Springer

Hyper-HeuristicsBurke EK, Hyde M, Kendall G, Ochoa G, Ozcan E, Woodward J

Handbook of Meta-heuristics, chap A Classification of Hyper-heuristics Approaches

Portfolios methodsGagliolo M, Schmidhuber J (2008) Algorithm selection as a bandit problem with unbounded losses. Tech. rep., Tech.

report IDSIA - 07

Why introducing the concept of Autonomous Search ?

Taxonomy

Taxonomies

Page 53: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Classification : Solving

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

On-

line

Off-

line Au

to

Complete/incomplete search,Model representationOther optimization paradigms (e.g., ACO )

Taxonomy

Page 54: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Classification : Parameters

Solving MethodsTree-Based Search SLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Numerical/discrete valuesComponents of the solverVs. Configuration of the solver

Taxonomy

Page 55: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Classification : Settings

Solving MethodsTree-Based Search EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

•Experiment-based•Feedback Control •Measures and learning techniques(reinforcement learning, statistical learning, case-base reasonning…)

Taxonomy

Page 56: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Related Approaches

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Parameter Setting in Evolutionary Computation

Page 57: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Related Approaches

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Parameter Setting

ParameterTuning

ParameterControl

Deterministic

Adaptive Self-adaptive

Page 58: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Related Approaches

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Optimization of algorithms (automated tuning)

Page 59: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Related Approaches

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Optimization of algorithms (automated tuning)

SLS Based (ParamILS)Frank Hutter, Holger H. Hoos, Kevin Leyton-Brown, Thomas Stützle: ParamILS: An Automatic Algorithm Configuration Framework. J. Artif. Intell. Res. (JAIR) 36: 267-306 (2009)

GA Based (Revac)Volker Nannen, A. E. Eiben: Efficient relevance estimation and value calibration of evolutionary algorithm parameters. IEEE Congress on Evolutionary Computation 2007: 103-110

Racing techniquesMauro Birattari, Thomas Stützle, Luis Paquete, Klaus Varrentrapp: A Racing Algorithm for Configuring Metaheuristics. GECCO 2002: 11-18

Page 60: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Related Approaches

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Reactive SearchLearning for SLS

Page 61: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Related Approaches

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Hyper heuristics

Page 62: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Hyper Heuristics

Taxonomy

Two possible views

• heuristics to choose heuristics

• heuristics to generate heuristics

Burke EK, Hyde M, Kendall G, Ochoa G, Ozcan E, Woodward JHandbook of Meta-heuristics, A Classification of Hyper-heuristicsApproaches

Page 63: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Proposing a general view of AS

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Taxonomy

Autonomous Search

Page 64: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Autonomous Search Genesis

Taxonomy

Gather works from the CSP solving community

Workshop “Autonomous Search” CP 2007, Providence (RI)

Page 65: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Autonomous Search Genesis

Taxonomy

Identify common concepts, goals andchallenges for future works

Page 66: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

Requirements for an Autonomous Search SystemParameters

Taxonomy

Modify its internal components

•Parameters•Fine grain heuristics•Coarse grain solving techniques•Model representation

React to external forces and opportunities

•Search landscape analysis (quality, diversity,...)•External knowledge (prediction models, rules, ...)

Page 67: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

CS Related Areas

Taxonomy

Solving Techniques Point of View

•Constraint Programming•Operation Research•Evolutionary Computation

Adjustment Techniques Point of View

•Reinforcement Learning•Statistical Learning•Information Theory

Page 68: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Tuning the parameters

CS Related Areas

Taxonomy

Solving Techniques Point of View

•Constraint Programming•Operation Research•Evolutionary Computation

Adjustment Techniques Point of View

•Reinforcement Learning•Statistical Learning•Information Theory

Not limited to…

Page 69: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioralExamples

Page 70: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Manual Empirical TuningDeciding the Size of a Tabu List

Mazure B, Sais L, Gregoire E, Tabu search for sat. In : AAAI/IAAI,pp 281285, 1997

Page 71: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Automated Parameter Tuning(SLS based)Hutter F, Hoos H, Stutzle T Automatic algorithm configuration based on local search. AAAI 2007

Page 72: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Portfolios ApproachesFeatures based linear regression and classiers

Xu L, Hutter F, Hoos HH, Leyton-Brown K Satzilla : portfolio-based algorithm selection for sat. JAIR 2008

Page 73: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Learning Combinations of Well-known Heuristics

Epstein S, Freuder E, Wallace R Learning to support constraint programmers. Comput Intell 2005

Page 74: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Discovering heuristics(variable selection in SAT SLS)

Alex S. Fukunaga : Automated Discovery of Local Search Heuristics forSatisability Testing. Evolutionary Computation 2008

Page 75: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Automated TuningAdjusting the size of a Tabu List

R. Battiti, G. Tecchiolli : The Reactive Tabu Search. INFORMS Journal on Computing 6(2): 126-140 (1994)

Page 76: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Controlling Variable Orderings and Values Selection inSearch Heuristics

Boussemart F, Hemery F, Lecoutre C, Sais L Boosting systematic search by weighting constraints. ECAI2004 2004

Page 77: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Examples of works

Examples

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Adaptive Genetic Algorithms

A.E. Eiben, Z. Michalewicz, M. Schoenauer, J. E. SmithParameter Control in Evolutionary Algorithms.In Parameter Setting in Evolutionary Algorithms

2007

Page 78: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example : how to design an AS system ?

Generic Evolutionary Algorithm for Constraint Satisfaction Problem

Focus

Population

Apply variation operator

New Population

Solution ?

Page 79: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

Generic Search Algorithm for ConstraintSatisfaction Problem

Focus

Configuration

Apply variation operator

New configuration

Solution ?

Question : How to choose the suitable operator at each iteration ?

Page 80: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

Idea

Associate a probability of application to each operator (parameter)

Select an operator according to thisprobability scheme

Focus

Page 81: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

Question : how to set the probabilities(parameters of the algorithm) ?

Focus

Page 82: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

Use a principled approach to tune yourparameter

•Search in the parameters space

•Assess the performance of each setting through runs of the algorithm on selected Instances

ParamILS, REVAC, F-Race …

Focus

Page 83: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

Second idea:

Control the probability during the run

•Evaluate the performance of each operator after application

•Adjust the parameters according to the performances

Focus

Page 84: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

General process for control(Automated Operator Selection)

Focus

Page 85: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

What are the suitable criteria ?

-Quality -Fitness diversity -Genotypic diversity -Time -…

Focus

Page 86: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

What are the suitable criteria ?

-Quality -Fitness diversity -Genotypic diversity -Time -…

Focus

Page 87: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Different performnce mesearues

Focus

1 2 3 4 5 6 70

2

4

6

8

10

12

14

16

Op1Op2

Sliding Windows

Mean or Max ?

How to measure the impact ?

Page 88: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

What is the performance of the operators ?

•Fix a search policy•Dynamic policy •Values against rank •…

Focus

Page 89: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

What is the performance of the operators ?

Fix a search policy

Focus

Page 90: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Focus on an example

What is the performance of the operators ?No values :

Pareto rank of the operatorsArea under the curve

Focus

Page 91: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Estimating efficience of operators

How to reward the operators ?

Proportionally to their performance

Focus

Page 92: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Estimating efficience of operators

Using UCB (Upper Confidence Bound)(reinforcement learning technique)

Exploration + Exploitation of the operators

Choosing the operator having the best UCB

Focus

to

ktk

to n

nCr

,

,

,

)log(

Page 93: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Estimating efficience of operators

Warning : UCB converge asymptotically toGain for the MAB

But here we have dynamic changes

Use of statistical test to restart learning.

Focus

Page 94: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Different selection processes

Focus

UCB PM Uniform0

2

4

6

8

10

12

14

16

18

Op1Op2Op3Op4

Page 95: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Instances ?

Comparisons ?

Induced new parameters ? ?

Reliability ?

How to assess the performances of your system ?

Focus

Page 96: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

What’s next ?

Solving MethodsTree-Based Search

MetaheuristicsSLS EA

Para

met

er s

etting

met

hod

On-

line

Off-

line Au

to

Parameter type

StructuralBehavioral

Focus

Page 97: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Conclusion

Many different possible approaches

Guidelines for designing new autonomousSolvers

Off-line/On-lineBehavioural parameters/componentsControl of the efficient heuristics/discovering new heuristics

Conclusion

Page 98: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Challenges

Comparing performances

•Autonomous vs. ad-hoc

•Off-line Tuning vs. On-line control

•Representative benchmarking

Conclusion

Page 99: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Challenges

Comparing performances

•Methodologies for comparisons

•New competitions Chesc (Cross-domain Heuristic Search Challenge)

G. Ochoa and her team

•Related to No Free-Lunch TheoremsMore reliable on more problems

Conclusion

Page 100: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Challenges

Parameters induced by the AS system

•Abstract parameters should be more easy to control (e.g., EvE balance)

•New parameters should be less sensitive than original ones

•Fewer paramaters are easier to adjust

Conclusion

Page 101: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Challenges

Learning

•Interactions solving-learning

•Improving learning off-line

•Short term (react) vs. long term (prediction)

•Continuous search (Arbelaez, Hamadi & Sebag)

Conclusion

Page 102: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Challenges

Distributed and parallel computing

•Improving algorithm’s space exploration

•Sharing information on parameters

•Sharing information on problems

Conclusion

Page 103: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Challenges

Towards more generic on-line control tools

•Identify generic control techniques andmeasures

•Control various components type (behavioral parameters, objective functions, heuristics…)

Conclusion

Page 104: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

Of course all LION Proceedings …

Some books to read

Conclusion

Page 105: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

And ;-)

Conclusion

Page 106: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

So

Sorry for missing references and works

Conclusion

I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references

Page 107: Frédéric Saubion LERIA Learning and Intelligent OptimizatioN Conference Autonomous Search.

So

Questions

Conclusion

I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references I will not forget important works and references