Скачать презентацию Hybrid Soft Computing Where Are We Going Piero Скачать презентацию Hybrid Soft Computing Where Are We Going Piero

8718d4085e913aee5a9c02e6230fd46e.ppt

  • Количество слайдов: 115

Hybrid Soft Computing: Where Are We Going? Piero P. Bonissone GE Corporate Research & Hybrid Soft Computing: Where Are We Going? Piero P. Bonissone GE Corporate Research & Development [email protected]. ge. com

Hybrid SC and EA - Outline • Soft Computing Overview SC Components: PR, FL, Hybrid SC and EA - Outline • Soft Computing Overview SC Components: PR, FL, NN, EA • Modeling with FL and EA • Hybrid SC Systems FLC Parameter Tuning by EA Parameter Setting • Conclusions August 24, 2000 ECAI 2000 2

Hybrid SC and EA - Outline • Soft Computing Overview SC Components: PR, FL, Hybrid SC and EA - Outline • Soft Computing Overview SC Components: PR, FL, NN, EA • Modeling with FL and EA • Hybrid SC Systems FLC Parameter Tuning by EA Parameter Setting • Conclusions August 24, 2000 ECAI 2000 3

Soft Computing • Soft Computing (SC): the symbiotic use of many emerging problem-solving disciplines. Soft Computing • Soft Computing (SC): the symbiotic use of many emerging problem-solving disciplines. • According to Prof. Zadeh: ". . . in contrast to traditional hard computing, soft computing exploits the tolerance for imprecision, uncertainty, and partial truth to achieve tractability, robustness, low solution -cost, and better rapport with reality” • Soft Computing Main Components: Approximate Reasoning: » Probabilistic Reasoning, Fuzzy Logic Search & Optimization: » Neural Networks, Evolutionary Algorithms August 24, 2000 ECAI 2000 4

Problem Solving Techniques HARD COMPUTING SOFT COMPUTING Precise Models Approximate Models Symbolic Logic Reasoning Problem Solving Techniques HARD COMPUTING SOFT COMPUTING Precise Models Approximate Models Symbolic Logic Reasoning August 24, 2000 Traditional Numerical Modeling and Search ECAI 2000 Approximate Reasoning Functional Approximation and Randomized Search 5

Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Approximation/ Randomized Search Neural Networks Evolutionary Algorithms Bayesian Belief Nets Dempster Shafer August 24, 2000 ECAI 2000 6

Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Approximation/ Randomized Search Neural Networks Evolutionary Algorithms Bayesian Belief Nets Dempster Shafer HYBRID PROBABILISTIC SYSTEMS Probability of Fuzzy Events August 24, 2000 Belief of Fuzzy Events ECAI 2000 Fuzzy Influence Diagrams 7

Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Approximation/ Randomized Search Neural Networks Evolutionary Algorithms Bayesian Belief Nets Dempster Shafer HYBRID PROBABILISTIC SYSTEMS Probability of Fuzzy Events August 24, 2000 Belief of Fuzzy Events ECAI 2000 Fuzzy Influence Diagrams 8

Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Soft Computing: Hybrid Probabilistic Systems Approximate Reasoning Probabilistic Models Multivalued & Fuzzy Logics Functional Approximation/ Randomized Search Neural Networks Evolutionary Algorithms Bayesian Belief Nets Dempster Shafer HYBRID PROBABILISTIC SYSTEMS Probability of Fuzzy Events August 24, 2000 Belief of Fuzzy Events ECAI 2000 Fuzzy Influence Diagrams 9

Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Fuzzy Systems Neural Networks Evolutionary Algorithms Multivalued Algebras Fuzzy Logic Controllers August 24, 2000 ECAI 2000 10

Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Fuzzy Systems Neural Networks Evolutionary Algorithms Multivalued Algebras Fuzzy Logic Controllers HYBRID FL SYSTEMS NN modified by FS (Fuzzy Neural Systems) August 24, 2000 FLC Tuned by NN (Neural Fuzzy Systems) ECAI 2000 FLC Generated and Tuned by EA 11

Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Fuzzy Systems Neural Networks Evolutionary Algorithms Multivalued Algebras Fuzzy Logic Controllers HYBRID FL SYSTEMS NN modified by FS (Fuzzy Neural Systems) August 24, 2000 FLC Tuned by NN (Neural Fuzzy Systems) ECAI 2000 FLC Generated and Tuned by EA 12

Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid FL Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Fuzzy Systems Neural Networks Evolutionary Algorithms Multivalued Algebras Fuzzy Logic Controllers HYBRID FL SYSTEMS NN modified by FS (Fuzzy Neural Systems) August 24, 2000 FLC Tuned by NN (Neural Fuzzy Systems) ECAI 2000 FLC Generated and Tuned by EA 13

Soft Computing: Hybrid NN Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid NN Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Neural Networks Feedforward NN RBF August 24, 2000 Recurrent NN Single/Multiple Layer Perceptron Hopfield Evolutionary Algorithms ECAI 2000 SOM ART 14

Soft Computing: Hybrid NN Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid NN Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Neural Networks Feedforward NN RBF Recurrent NN Single/Multiple Layer Perceptron Hopfield HYBRID NN SYSTEMS NN parameters (learning rate h momentum a ) controlled by FLC August 24, 2000 Evolutionary Algorithms ECAI 2000 SOM ART NN topology &/or weights generated by EAs 15

Soft Computing: Hybrid NN Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid NN Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Neural Networks Feedforward NN RBF Recurrent NN Single/Multiple Layer Perceptron Hopfield HYBRID NN SYSTEMS NN parameters (learning rate h momentum a ) controlled by FLC August 24, 2000 Evolutionary Algorithms ECAI 2000 SOM ART NN topology &/or weights generated by EAs 16

Soft Computing: Hybrid EA Systems Approximate Reasoning Probabilistic Multivalued & Fuzzy Logics Models Functional Soft Computing: Hybrid EA Systems Approximate Reasoning Probabilistic Multivalued & Fuzzy Logics Models Functional Approximation/ Randomized Search Neural Networks Evolution Strategies Evolutionary Programs August 24, 2000 ECAI 2000 Evolutionary Algorithms Genetic Progr. 17

Soft Computing: Hybrid EA Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid EA Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Neural Networks Evolution Strategies Evolutionary Programs Evolutionary Algorithms Genetic Progr. HYBRID EA SYSTEMS EA parameters (N, P ) cr mu EA parameters (Pop size, select. ) controlled by FLC August 24, 2000 EA-based search inter-twined with hill-climbing controlled by EA ECAI 2000 18

Soft Computing: Hybrid EA Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid EA Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Neural Networks Evolution Strategies Evolutionary Programs Evolutionary Algorithms Genetic Progr. HYBRID EA SYSTEMS EA parameters (N, P ) cr mu EA parameters (Pop size, select. ) controlled by FLC August 24, 2000 EA-based search inter-twined with hill-climbing controlled by EA ECAI 2000 19

Soft Computing: Hybrid EA Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Soft Computing: Hybrid EA Systems Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Neural Networks Evolution Strategies Evolutionary Programs Evolutionary Algorithms Genetic Progr. HYBRID EA SYSTEMS EA parameters (N, P ) cr mu EA parameters (Pop size, select. ) controlled by FLC August 24, 2000 EA-based search inter-twined with hill-climbing controlled by EA ECAI 2000 20

Hybrid SC and EA – Outline (2) • Soft Computing Overview SC Components: PR, Hybrid SC and EA – Outline (2) • Soft Computing Overview SC Components: PR, FL, NN, EA • Modeling with FL and EA • Hybrid SC Systems FLC Parameter Tuning by EA Parameter Setting • Conclusions August 24, 2000 ECAI 2000 21

Fuzzy Logic Genealogy • Origins: MVL for treatment of imprecision and vagueness 1930 s: Fuzzy Logic Genealogy • Origins: MVL for treatment of imprecision and vagueness 1930 s: Post, Kleene, and Lukasiewicz attempted to represent undetermined, unknown, and other possible intermediate truth-values. 1937: Max Black suggested the use of a consistency profile to represent vague (ambiguous) concepts 1965: Zadeh proposed a complete theory of fuzzy sets (and its isomorphic fuzzy logic), to represent and manipulate ill-defined concepts August 24, 2000 ECAI 2000 22

Fuzzy Logic : Linguistic Variables • Fuzzy logic give us a language (with syntax Fuzzy Logic : Linguistic Variables • Fuzzy logic give us a language (with syntax and local semantics), in which we can translate our qualitative domain knowledge. • Linguistic variables to model dynamic systems • These variables take linguistic values that are characterized by: a label - a sentence generated from the syntax a meaning - a membership function determined by a local semantic procedure August 24, 2000 ECAI 2000 23

Fuzzy Logic : Reasoning Methods • The meaning of a linguistic variable may be Fuzzy Logic : Reasoning Methods • The meaning of a linguistic variable may be interpreted as a elastic constraint on its value. • These constraints are propagated by fuzzy inference operations, based on the generalized modus-ponens. • A FL Controller (FLC) applies this reasoning system to a Knowledge Base (KB) containing the problem domain heuristics. • The inference is the result of interpolating among the outputs of all relevant rules. • The outcome is a membership distribution on the output space, which is defuzzified to produce a crisp output. August 24, 2000 ECAI 2000 24

Fuzzy Logic Control : Inference Method State Variables Output Variable S LN S L Fuzzy Logic Control : Inference Method State Variables Output Variable S LN S L SN L S SP L Rules S L LP Interpolation Defuzzification August 24, 2000 Inputs ECAI 2000 25

FLC Inference Method (cont. ) • A FLC (KB + Reasoning Mechanism) defines a FLC Inference Method (cont. ) • A FLC (KB + Reasoning Mechanism) defines a deterministic response surface in the cross product of state and output spaces, which approximates the original relationship. • The FLC leverages the interpolation properties of this reasoning mechanism, to exhibit robustness with respect to parameter variations, disturbances, etc. August 24, 2000 ECAI 2000 26

Example (MISO): Max-min Composition with Centroid Defuzzification Rules set • If X is SMALL Example (MISO): Max-min Composition with Centroid Defuzzification Rules set • If X is SMALL and Y is SMALL then Z is NEG. LARGE • If X is SMALL and Y is LARGE the Z is NEG. SMALL • If X is LARGE and Y is SMALL the Z is POS. SMALL • If X is LARGE and Y is LARGE then Z is POS. LARGE Response Surface Term set August 24, 2000 ECAI 2000 27

Evolutionary Algorithms (EA) EA are part of the Derivative-Free Optimization and Search Methods: Evolutionary Evolutionary Algorithms (EA) EA are part of the Derivative-Free Optimization and Search Methods: Evolutionary Algorithms Simulated annealing (SA) Random search Downhill simplex search Tabu search EA consists of: Evolution Strategies (ES) Evolutionary Programming (EP) Genetic Algorithms (GA) Genetic Programming (GP) August 24, 2000 ECAI 2000 28

Evolutionary Algorithms Characteristics • Most Evolutionary Algorithms can be described by x[t + 1] Evolutionary Algorithms Characteristics • Most Evolutionary Algorithms can be described by x[t + 1] = s(v(x[t])) - x[t] : the population at time t under representation x - v : is the variation operator(s) - s : is the selection operator August 24, 2000 ECAI 2000 29

Evolutionary Algorithms Characteristics • EA exhibit an adaptive behavior that allows them to handle Evolutionary Algorithms Characteristics • EA exhibit an adaptive behavior that allows them to handle non linear, high dimensional problems without requiring differentiability or explicit knowledge of the problem structure. • EA are very robust to time varying behavior, even though they may exhibit low speed of convergence. August 24, 2000 ECAI 2000 30

Modeling • Model = Structure + Parameters + Search Method • Classical control theory: Modeling • Model = Structure + Parameters + Search Method • Classical control theory: Structure: order of the differential equations Parameters: coefficients of differential equation. Search method: LMSE, Pole-placement, etc. August 24, 2000 ECAI 2000 31

Modeling Using FLC (Mamdani type) • A Mamdani- type FLC approximates a relationship between Modeling Using FLC (Mamdani type) • A Mamdani- type FLC approximates a relationship between a state X and an output Y by using a KB and a reasoning mechanism (generalized modus-ponens). • The Knowledge Base (KB) is defined by: Scaling factors (SF): ranges of values of state and output variables Termset (TS): membership functions of values Ruleset (RS): a syntactic mapping of symbols from X to Y August 24, 2000 ECAI 2000 32

Modeling Using FLC (Mamdani type) • The structure of the model is the ruleset. Modeling Using FLC (Mamdani type) • The structure of the model is the ruleset. • The parameters of the model are the scaling factors and termsets. • The search method is initialized by knowledge engineering and refined with some other external methods (SOFC, error minimization, etc. ) August 24, 2000 ECAI 2000 33

Modeling Using EA • Similarly, for EA: The structure of the model is the Modeling Using EA • Similarly, for EA: The structure of the model is the representation of an individual in the population (e. g. , binary string, vector, parse tree, Finite State Machine). The parameters of the model are the Population Size, Probability of Mutation, Prob. of Recombination, Generation Gap, etc. The search method is a global search based on maximization of population fitness function August 24, 2000 ECAI 2000 34

Hybrid SC and EA – Outline (3 a) • Soft Computing Overview SC Components: Hybrid SC and EA – Outline (3 a) • Soft Computing Overview SC Components: PR, FL, NN, EA • Modeling with FL and EA • Hybrid SC Systems FLC Parameter Tuning by EA Parameter Setting • Conclusions August 24, 2000 ECAI 2000 35

Hybrid Soft Computing: FLC Tuned by EAs Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Hybrid Soft Computing: FLC Tuned by EAs Approximate Reasoning Functional Approximation/ Randomized Search Probabilistic Multivalued & Fuzzy Logics Models Fuzzy Systems Fuzzy Logic Controllers Neural Networks Multivalued Algebras Evolution Strategies Evolutionary Programs Evolutionary Algorithms Genetic Progr. HYBRID FLC/EA SYSTEMS FLC Generated and Tuned by EA August 24, 2000 ECAI 2000 36

FLC Tuned by EA - Outline • Components & Historical Approaches • Application to FLC Tuned by EA - Outline • Components & Historical Approaches • Application to Automatic Train Handling (ATH) • Solution Architecture • Analysis of Results • Remarks August 24, 2000 ECAI 2000 37

FL Controllers Tuned by EAs • FLC = KB + Inference Engine (with Defuzz. FL Controllers Tuned by EAs • FLC = KB + Inference Engine (with Defuzz. ) KB parameters: » Scaling factors (SF) » Membership Functions (MF) » Rule set (RS) • EA August 24, 2000 Encoding: binary or real valued Chromosome: string or table Fitness function: Sum quadratic errors, entropy Operators: one point crossover, max min arithmetical crossover, point radius crossover. ECAI 2000 38

FL Controllers tuned by EAs (cont. ) • Historical Approaches: Karr 91 -93: » FL Controllers tuned by EAs (cont. ) • Historical Approaches: Karr 91 -93: » Chromosome = concatenation of all termsets. » Each value in a termset was represented by 3 binary encoded parameters. Lee & Takagi 93: » Chromosome = 1 TSK rule (LHS: memb. fnct. RHS pol. ) » Binary encoding of 3 parameter repr. of each term Surman et al: 93: » Fitness function with added entropy term describing number of activated rules August 24, 2000 ECAI 2000 39

SC in Train Handling: An Example • Problem Description Develop an automated train handler SC in Train Handling: An Example • Problem Description Develop an automated train handler to control a massive, distributed system with little sensor information Freight trains consist of several hundred heavy railcars connected by couplers (train length up to two miles) Each coupler typically has a dead zone and a hydraulically damped spring Railcars can move relative to each other while in motion, leading to a train that can change its length by 50 – 100 ft. The position of the cars and couplers cannot be electronically sensed August 24, 2000 ECAI 2000 40

SC in Train Handling: An Example • Solution Requirements • An automated system has SC in Train Handling: An Example • Solution Requirements • An automated system has to satisfy multiple goals: Tracking a velocity reference (defined over distance) to enforce speed limits and respect the train schedule Providing a degree of train handling uniformity across all crews Operating the train in fuel efficient regimes Maintaining a smooth ride by avoiding sudden accelerations or brake applications (slack control) Multi-body regulation problem, subject to proper slack management, without sensors for most of the state August 24, 2000 ECAI 2000 41

SC in Train Handling: An Example • Description of Our Approach Use a Velocity SC in Train Handling: An Example • Description of Our Approach Use a Velocity Profile externally generated (using classical optimization or Evolutionary Algorithms) Use a Fuzzy Logic Control (FLC) to track the velocity reference (Fuzzy PI Control) Use an Evolutionary Algorithms to tune the FLC parameters to minimize velocity tracking error and number of throttle changes Implement control actions with fuzzy rule set to maintain slack control August 24, 2000 ECAI 2000 42

FLC tuned by EAs: Our Approach • Chromosome (real-valued encoding) Chr. 1 = Scaling FLC tuned by EAs: Our Approach • Chromosome (real-valued encoding) Chr. 1 = Scaling factors; Chr. 2 = Termsets; Chr. 3 = Rules (not used) • Order of tuning (as in Zheng '92): Initialize rulebase with standard PI structure and termsets with uniformly distributed terms Apply EAs to find best scaling factors Apply EAs to find best termsets Apply EAs to find best rule set (not used) • Transition from large to small granularity August 24, 2000 ECAI 2000 43

FLC Sensitivity to Parameter Changes Changing a Scaling Factor Changing a Term in X FLC Sensitivity to Parameter Changes Changing a Scaling Factor Changing a Term in X 1 Changing a Rule August 24, 2000 ECAI 2000 44

Architecture: Modules, Fitness Funct. • Architecture EA: pop. size=50; P(cross)=. 6; P(mut)=. 001 Three Architecture: Modules, Fitness Funct. • Architecture EA: pop. size=50; P(cross)=. 6; P(mut)=. 001 Three Types of fitness functions Train Simulator: NSTD (STD+TEM) Fuzzy PI (Ke, Kedot, K u) • Fitness functions (f 1, f 2, f 3) August 24, 2000 ECAI 2000 45

FLC tuned by GAs SF or MF GA (GENESIS) Fitness Function Train Simulator August FLC tuned by GAs SF or MF GA (GENESIS) Fitness Function Train Simulator August 24, 2000 FLC (PI) ECAI 2000 46

Experiment Design • 12 test (4 for each fitness function) Initial SF with initial Experiment Design • 12 test (4 for each fitness function) Initial SF with initial MF; EA tuned SF with Initial MF Initial SF with EA tuned MF; EA tuned SF with EA tuned MF • Train Simulation: 14 miles long flat track 1 uniformly heavy train with 100 cars and 4 locomotives Analytically computed velocity profile August 24, 2000 ECAI 2000 47

Experiment Design • Representation: SF: 3 floating point values for Ke, Kedot, KDu MF Experiment Design • Representation: SF: 3 floating point values for Ke, Kedot, KDu MF (21 -9) = 12 values » 21 parameters: [(Lefti , Centeri , Righti ) for i=1, . . . , 7] » 9 dependent values: [(Lefti = Right(i+1)) for i=1, . . . , 6] + [Center 1= Center 7]+[Right 1 = Left 7 = 0] Constraints to maintain 0. 5 terms overlap, for best interpolation August 24, 2000 ECAI 2000 48

Experiments Results • Experiment Results with f 1 • Experiment Results with f 3 Experiments Results • Experiment Results with f 1 • Experiment Results with f 3 August 24, 2000 ECAI 2000 49

Tuning of FLC with EA: Remarks • Verified tuning order proposed by Zheng (92) Tuning of FLC with EA: Remarks • Verified tuning order proposed by Zheng (92) » SF tuning: major impact » MF tuning: minor impact » RS tuning: almost no impact • For both f 1 and f 3, fuel minimization is implicitly derived from throttle jockeying minimization • Complex fitness function (requiring simulation run 23 sec for each chromosome evaluation) limited trials number with no apparent impact • Successfully tested on simulated 43 mile long track with altitude excursions » (Selkirk, NY >Framingham, MA) August 24, 2000 ECAI 2000 50

Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 40. 00 30. 00 20. 00 10. 00 NOTCH POSITION mile NOTCH POSITION 8 4 August 24, 2000 ECAI 2000 0 mile 51

Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 40. 00 30. 00 20. 00 10. 00 NOTCH POSITION mile NOTCH POSITION 8 4 August 24, 2000 ECAI 2000 0 mile 52

Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 40. 00 30. 00 20. 00 10. 00 NOTCH POSITION mile NOTCH POSITION 8 4 August 24, 2000 ECAI 2000 0 mile 53

Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 Results of EA Tuned PI on 43 mile Track MPH 60. 00 50. 00 40. 00 30. 00 20. 00 10. 00 NOTCH POSITION mile NOTCH POSITION 8 4 August 24, 2000 ECAI 2000 0 mile 54

Hybrid SC and EA – Outline (3 b) • Soft Computing Overview SC Components: Hybrid SC and EA – Outline (3 b) • Soft Computing Overview SC Components: PR, FL, NN, EA • Modeling with FL and EA • Hybrid SC Systems FLC Parameter Tuning by EA Parameter Setting • Conclusions August 24, 2000 ECAI 2000 55

EA Parameter Setting • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter EA Parameter Setting • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Tuning EA Parameter Control • An Application to Agile Manufacturing Object level Representation and Complexity • Solution FLC KB Statistical Experiments Analysis and Summary of 1200 Experiments • Remarks August 24, 2000 ECAI 2000 56

EA Model Object-level GA Structure & Parameters Object level Problem August 24, 2000 ECAI EA Model Object-level GA Structure & Parameters Object level Problem August 24, 2000 ECAI 2000 57

EA Structure • GA Structural Design Selections: GA Type: » {Simple, Steady State, Niche, EA Structure • GA Structural Design Selections: GA Type: » {Simple, Steady State, Niche, …} Chromosome Encoding: » {Binary, Integer, Real, . . . } Constraints Representation: » {Penalty function, data structure, filters, …} Fitness Function: » {Scalar function, Weighted aggregation of multiple functions, Vector valued function, …} August 24, 2000 ECAI 2000 58

EA Parameters • Adjustable parameters for a GA N = Population size Large pop. EA Parameters • Adjustable parameters for a GA N = Population size Large pop. prevent premature convergence Pc = Crossover rate: Pcr * N = # crossovers per generation Pm = Mutation rate: Pm * N * L = # mutations per generation G = Generation Gap Percentage of population to be replaced W = Scaling Window Size =[1, 7] S = Selection Strategy = {Elitist, Non Elitist} August 24, 2000 ECAI 2000 59

EA Parameter Setting - Outline • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Setting - Outline • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Tuning EA Parameter Control • An Application to Agile Manufacturing Object level Representation and Complexity • Solution FLC KB Statistical Experiments Analysis and Summary of 1200 Experiments • Remarks August 24, 2000 ECAI 2000 60

EAs Parameter Setting Before the run Parameter Setting Parameter Control Parameter Tuning August 24, EAs Parameter Setting Before the run Parameter Setting Parameter Control Parameter Tuning August 24, 2000 During the run ECAI 2000 61

EAs Parameter Setting Before the run Parameter Setting Parameter Control Parameter Tuning Deterministic August EAs Parameter Setting Before the run Parameter Setting Parameter Control Parameter Tuning Deterministic August 24, 2000 During the run ECAI 2000 Adaptive Self Adaptive 62

EAs Parameter Setting: Parameter Tuning Before the run Parameter Setting Parameter Control Parameter Tuning EAs Parameter Setting: Parameter Tuning Before the run Parameter Setting Parameter Control Parameter Tuning Deterministic August 24, 2000 During the run Adaptive Self Adaptive ECAI 2000 Off-line Tuning Determined before running the GAs on the object-level problem by » Studying a subset of five diverse problems (De. Jong, 1975) » Running a Meta Genetic Algorithm (Grefenstette, 1986) 63

Off Line Tuning of GA Parameters (De. Jong, 1975) Object-level GA Suite of 5 Off Line Tuning of GA Parameters (De. Jong, 1975) Object-level GA Suite of 5 problems: - Parabola - Rosenbrock’s saddle - Step function - Quartic Noise - Shekel’s foxholes Population Size: Crossover Rate: Mutation Rate: Replacement Scaling Window Selection Strategy 50 0. 6 0. 001 100% n=inf Elitist Object-level GA Object-level Problem August 24, 2000 ECAI 2000 64

SC Hybrid Systems: EA Tuning EA Approximate Reasoning Approaches Probabilistic Multivalued & Fuzzy Logics SC Hybrid Systems: EA Tuning EA Approximate Reasoning Approaches Probabilistic Multivalued & Fuzzy Logics Models Search/Optimization Approaches Neural Networks Evolutionary Algorithms Evolution Strategies Evolutionary Programs Genetic Algorithms Genetic Progr. HYBRID EA SYSTEMS EA parameters (Pop size, select. ) controlled by EA August 24, 2000 ECAI 2000 65

Off Line Tuning of GA Parameters (Grefenstette, 1986) Object GA Performance Meta- GA Object-level Off Line Tuning of GA Parameters (Grefenstette, 1986) Object GA Performance Meta- GA Object-level GA Suite of 5 problems: - Parabola - Rosenbrock’s saddle - Step function - Quartic Noise - Shekel’s foxholes Object GA Parameter Set Off-Line Performance Population Size: 80 On-Line Performance Crossover Rate: 0. 45 Mutation Rate: Population Size: Replacement Crossover Rate: Scaling Window Mutation Rate: Selection Strategy Replacement Scaling Window Selection Strategy 0. 01 30 90% 0. 96 n= 0. 01 1 Non. Elitist 100% n = inf Elitist Object-level GA Object-level Problem August 24, 2000 ECAI 2000 66

GAs Parameter Setting: Deterministic Control Before the run Parameter Tuning Deterministic August 24, 2000 GAs Parameter Setting: Deterministic Control Before the run Parameter Tuning Deterministic August 24, 2000 During the run Parameter Setting Parameter Control Adaptive Self Adaptive ECAI 2000 No feedback information is used. A time-varying schedule is used to modify a GA parameter p p is replaced by p(t) Correct design of p(t) is very difficult 67

EAs Parameter Setting: Deterministic Control - Example Control of Population size By decreasing Population EAs Parameter Setting: Deterministic Control - Example Control of Population size By decreasing Population Size toward the last part of the Evolution we are trying to improve the solution refinement (e. g. , more generations with same number of trials) • Constant Population size: N = 338 • Number of trials = 338 * Max. Gen • Variable Population size: N(t) • Number of trials = 338 * Max. Gen August 24, 2000 ECAI 2000 68

EAs Parameter Setting: Self-Adaptive Control Before the run Parameter Setting During the run Incorporate EAs Parameter Setting: Self-Adaptive Control Before the run Parameter Setting During the run Incorporate parameters into chromosome making them subject to evolution Typically used to determine Mutation Step S: [g 1 g 2. . . gn S] Parameter Control Parameter Tuning Mutation Step for Entire Genome Deterministic Adaptive or Self. Adaptive [g 1 g 2. . . gn S 1 S 2. . . Sn ] Mutation Steps for Each Genome Value August 24, 2000 ECAI 2000 69

GAs Parameter Setting: Adaptive Control Before the run Parameter Setting During the run Parameter GAs Parameter Setting: Adaptive Control Before the run Parameter Setting During the run Parameter Control Parameter Tuning Deterministic Adaptive Feedback from the search is used to determine the direction and/or magnitude of the change in the parameter value. A Fuzzy Logic Controller is used to obtain parameter changes in Self Adaptive » Population Size » Mutation Rate as a function of » Genotypic Diversity » Percentage Completed Trials August 24, 2000 ECAI 2000 70

SC Hybrid Systems: FLC Tuning EA Search/Optimization Approaches Approximate Reasoning Approaches Probabilistic Multivalued & SC Hybrid Systems: FLC Tuning EA Search/Optimization Approaches Approximate Reasoning Approaches Probabilistic Multivalued & Fuzzy Logics Models MV-Algebras Neural Networks Fuzzy Logic Fuzzy Controller Evolution Strategies Genetic Algorithms Evolutionary Programs Genetic Progr. HYBRID SYSTEMS August 24, 2000 Evolutionary Algorithms ECAI 2000 EA parameters controlled by FLC 71

Fuzzy Logic Controlled GA (FLC-GA) State Variables describing the evolution stage • Genotypic Diversity Fuzzy Logic Controlled GA (FLC-GA) State Variables describing the evolution stage • Genotypic Diversity • Percentage Completed Trials Controlled GA parameters KB Fuzzy Logic Controller D Population Size D Mutation Rate Object-level GA Object level Problem August 24, 2000 ECAI 2000 72

EA Parameter Setting • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter EA Parameter Setting • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Tuning EA Parameter Control • An Application to Agile Manufacturing Object level Representation and Complexity • Solution FLC KB Statistical Experiments Analysis and Summary of 1200 Experiments • Remarks August 24, 2000 ECAI 2000 73

EA Parameter Control: An Application Global optimization of design, manufacturing, supplier planning decisions in EA Parameter Control: An Application Global optimization of design, manufacturing, supplier planning decisions in a distributed manufacturing environment Marketing Customer Tools Data Tools Design Data Tools Data Supplier Data Tools Virtual Design Environment Manufacturing August 24, 2000 ECAI 2000 74

Object-level Problem Representation Parts, suppliers, and design DB Part P 1 Part P 2 Object-level Problem Representation Parts, suppliers, and design DB Part P 1 Part P 2 Part Pk Acceptable Alternates Gene Allele Sets Design 8 6 3 3 2 2 1 1 Mfg. DB Suppliers P 1 P 2 Pk M Genome Manufacturing Facilities Offspring min i, j Object-level Optimization Problem August 24, 2000 Parents Crossover Operation ECAI 2000 Mutation 75

Object-level Problem Complexity Search Space Size • For EA Statistical Analysis: O(107) • For Object-level Problem Complexity Search Space Size • For EA Statistical Analysis: O(107) • For EA Performance Validation: O(1018) and O(1021) August 24, 2000 ECAI 2000 76

EA Parameter Setting - Outline • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Setting - Outline • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Tuning EA Parameter Control • An Application to Agile Manufacturing Object level Representation and Complexity • Solution FLC KB Statistical Experiments Analysis and Summary of 1200 Experiments • Remarks August 24, 2000 ECAI 2000 77

Solution Architecture Fuzzy Logic Controlled GA (Online Control) Fuzzy Logic Controller Untuned GA Object Solution Architecture Fuzzy Logic Controlled GA (Online Control) Fuzzy Logic Controller Untuned GA Object level GA Manufacturing Planning Module August 24, 2000 ECAI 2000 78

Untuned GA (U-TGA) Population Size: Generations: Crossover Rate: Mutation Rate: 50 250 0. 6 Untuned GA (U-TGA) Population Size: Generations: Crossover Rate: Mutation Rate: 50 250 0. 6 0. 001 Object level GA Manufacturing Planning Module August 24, 2000 ECAI 2000 79

Guidance for Experiments • Minimize high-level search space size for FLC-EA by - Identify Guidance for Experiments • Minimize high-level search space size for FLC-EA by - Identify primary drivers (influences) of EA search DOE determined that the two main drivers were: Population Size (N) and Mutation Rate (Pm ) - Control primary drivers by few simple heuristic rules Built two FLC controllers with heuristic rule sets and SF Changed on input (state variable) to capture evolution stage • Determining FLC firing rate - Take a control action every 10 generation • Extensive & statistically significant empirical evidence - Use t-test and F-tests to analyze m and s improvements August 24, 2000 ECAI 2000 80

Fuzzy Logic Controller for EAs: Knowledge Base I/O Scaling Factors I/O Termsets Rule Sets Fuzzy Logic Controller for EAs: Knowledge Base I/O Scaling Factors I/O Termsets Rule Sets KB • Genotypic Diversity • Percentage Completed Trials Fuzzy Logic Controller D Population Size D Mutation Rate Object-level EA Object-level Problem August 24, 2000 ECAI 2000 81

Fuzzy Controller for DN and DPm: Inputs • Inputs GD = Genotypic Diversity: Normalized Fuzzy Controller for DN and DPm: Inputs • Inputs GD = Genotypic Diversity: Normalized Average Hamming Distance where dij is the Hamming Distance GD range is [0, 1] == [Low, High] PFE = Percentage Fitness Evaluations: (Completed # Trials) / (Max Allocated # Trials) PFE range is [0, 1] = [Low, High] August 24, 2000 ECAI 2000 82

Fuzzy Logic Controller for EAs: Knowledge Base I/O Scaling Factors I/O Termsets Rule Sets Fuzzy Logic Controller for EAs: Knowledge Base I/O Scaling Factors I/O Termsets Rule Sets KB • Genotypic Diversity • Percentage Completed Trials Fuzzy Logic Controller D Population Size D Mutation Rate Object-level EA Object-level Problem August 24, 2000 ECAI 2000 83

Fuzzy Controller for DN and DPm: Termsets • Inputs: • High) GD: A(Very Low), Fuzzy Controller for DN and DPm: Termsets • Inputs: • High) GD: A(Very Low), B(Low), C(Medium), D(High), E(Very PFE: A(Very Low), B(Low), C(Medium), D(High), E(Very • Outputs (for both D N and 2000 Pm): D ECAI August 24, 2000 84

Fuzzy Logic Controller for EAs: Knowledge Base I/O Scaling Factors I/O Termsets Rule Sets Fuzzy Logic Controller for EAs: Knowledge Base I/O Scaling Factors I/O Termsets Rule Sets KB • Genotypic Diversity • Percentage Completed Trials Fuzzy Logic Controller D Population Size D Mutation Rate Object-level EA Object-level Problem August 24, 2000 ECAI 2000 85

Fuzzy Controller for Population Size: Rule Set GD, PFE->DN GD PFE DN = Genotypic Fuzzy Controller for Population Size: Rule Set GD, PFE->DN GD PFE DN = Genotypic Diversity: Normalized Average Hamming Distance = Percentage Fitness Evaluations: (Completed # Trials) / (Max Allocated # Trials) = Change in Population Size Exploration Stage Increase population/ broaden search August 24, 2000 ECAI 2000 Exploitation Stage Reduce population/ Refine best 86

Statistical Experiments: EA Structure • Data Set for Experiments Seven part classes corresponding to Statistical Experiments: EA Structure • Data Set for Experiments Seven part classes corresponding to a complexity of O(107) • EA Structure: Type: {Simple, Steady-State} Chromosome Encoding: Integer Fitness Function: Three type of cost functions Selection Method: Proportional Roulette Crossover Operator: Uniform August 24, 2000 ECAI 2000 87

Statistical Experiments: Set-Up • Set-Up for 1200 experiments: We defined 4 EA configurations (a) Statistical Experiments: Set-Up • Set-Up for 1200 experiments: We defined 4 EA configurations (a) Untuned Simple EA (U-SEA) (b) FL Controlled Simple EA (FLC-SEA) (c) Untuned Steady State EA (U-SSEA) (d) FL Controlled Steady State EA (FLCSSEA) August 24, 2000 ECAI 2000 88

Statistical Experiments: Set-Up (cont. ) • For each configuration we performed 300 experiments: 20 Statistical Experiments: Set-Up (cont. ) • For each configuration we performed 300 experiments: 20 runs for each pair of (Cost function, Max number of Trials) 15 different pairs of (Cost function, Max number of Trials) Three types of cost functions: (1) J = C*T; (2) J = C*T 2; (3) J = C*e(T-10)/3 Five values of maximum number of Trials (to evaluate effect of different evolution lengths): (i) 3, 000; 11, 000 August 24, 2000 (ii) 5, 000; (iii) 7, 000; ECAI 2000 (iv) 9, 000; (v) 89

Statistical Experiments: Measures • For each of the four configurations (a-d) we ran 20 Statistical Experiments: Measures • For each of the four configurations (a-d) we ran 20 experiments with the same parameters • Then we considered the following measures: B = sample average over 20 experiments of Best score frequency (number of time cost function J reached its minimal value known a priori for small size experiment) m = average of population best s = standard deviation of population best August 24, 2000 ECAI 2000 90

Statistical Experiments: Analysis • We performed an ANOVA test (both t and F test Statistical Experiments: Analysis • We performed an ANOVA test (both t and F test - with p < 0. 05 ) to see if: Cost (U-SEA) >> cost ( FLCSEA) Cost (U-SEA) >> cost ( U-SSEA) Cost (U-SSEA) >> cost ( FLC-SSEA) • We verified if the FLC caused the controlled EA to perform worse than its corresponding untuned EA, i. e. : Cost (U-SEA) << cost ( FLC-SEA) Cost (U-SSEA) << cost ( FLC-SSEA) August 24, 2000 ECAI 2000 91

Statistical Experiments: Results C*T 2 C*e(T-10)/3 Total Significant and changes in m in s Statistical Experiments: Results C*T 2 C*e(T-10)/3 Total Significant and changes in m in s Significant changes in s • For each cost function we ran 400 experiments (100 x EA type) • For each EA type we ran 20 experiments for 5 different pop. sizes • The entry in each cell is the number of significant changes found in the statistics of each of these five groups of experiments August 24, 2000 ECAI 2000 92

EA Parameter Setting • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter EA Parameter Setting • EA Model: Structure, Parameters • EA Parameter Setting EA Parameter Tuning EA Parameter Control • An Application to Agile Manufacturing Object level Representation and Complexity • Solution FLC KB Statistical Experiments Analysis and Summary of 1200 Experiments • Remarks August 24, 2000 ECAI 2000 93

Remarks • FLC State Representation: [Evolution Stage] Evolution time needs to be an explicit Remarks • FLC State Representation: [Evolution Stage] Evolution time needs to be an explicit state variable since we have different control goals during the EA’s stages. Diversity measures the evolutionary stage: » Percentage Fitness Evaluations (PFE) » Genotypic Diversity (GD) • FLC Control Variables: [EA Adaptable Parameters] - DN = Change in Population Size - DPm = Change in Mutation Rate August 24, 2000 ECAI 2000 94

Remarks (cont. ) • Main Result By using the FLC with the above State Remarks (cont. ) • Main Result By using the FLC with the above State and Control variables, we achieved a good improvement of the population average and an even better improvement of the population variance. No major negative effects on EA performance using FLC August 24, 2000 ECAI 2000 95

Hybrid SC and EA – Outline (4) • Soft Computing Overview SC Components: PR, Hybrid SC and EA – Outline (4) • Soft Computing Overview SC Components: PR, FL, NN, EA • Modeling with FL and EA • Hybrid SC Systems FLC Parameter Tuning by EA Parameter Setting • Conclusions August 24, 2000 ECAI 2000 96

Synergy in SC: Reasons & Approaches • Hybrid Soft Computing Leverages tolerance for imprecision, Synergy in SC: Reasons & Approaches • Hybrid Soft Computing Leverages tolerance for imprecision, uncertainty, and incompleteness intrinsic to the problems to be solved Generates tractable, low-cost, robust solutions to such problems by integrating knowledge and data • Tight Hybridization Data driven Tuning of Knowledge derived Models » Translate domain knowledge into initial structure and parameters » Use Global or local data search to tune parameters Knowledge driven Search Control » Use Global or local data search to derive models (Structure + Parameters) » Translate domain knowledge into an algorithm’s controller to improve/manage solution convergence and quality August 24, 2000 ECAI 2000 97

Synergy in SC: Reasons & Approaches • Loose Hybridization (Model Fusion) Does not combine Synergy in SC: Reasons & Approaches • Loose Hybridization (Model Fusion) Does not combine features of methodologies only their results Their outputs are compared, contrasted, and aggregated, to increase reliability • Hybrid Search Methods Intertwining local search within global search Embedding knowledge in operators for global search • Future: Circle of SC's related technologies will probably widen beyond its current constituents. Push for low cost solutions and intelligent tools will result in deployment of hybrid SC systems that efficiently integrate reasoning and search techniques. August 24, 2000 ECAI 2000 98

August 24, 2000 ECAI 2000 99 August 24, 2000 ECAI 2000 99

FL Controllers tuned by EAs (cont. ) • Historical Approaches (cont. ): Kinzel et FL Controllers tuned by EAs (cont. ) • Historical Approaches (cont. ): Kinzel et al. 94: » Chromosome = Rule Table » Point radius crossover changing 3 x 3 rule window (similar to a two point crossover for string representation) » Order of tuning: – Initialize rulebase according to heuristics – Apply GAs to find best rule table – Tune membership function of best rule set Herrera et al. 95: » Chromosome = concatenation of all rules » Real valued encoding, Max min arithmetical crossover August 24, 2000 ECAI 2000 100

Evolutionary Algorithms: ES • Evolutionary Strategies (ES) Originally proposed for the optimization of continuous Evolutionary Algorithms: ES • Evolutionary Strategies (ES) Originally proposed for the optimization of continuous functions (m , l)-ES and (m + l)-ES » A population of m parents generate l offspring » Best m offspring are selected in the next generation » (m , l) ES: parents are excluded from selection » (m + l) ES: parents are included in selection Started as (1+1)-ES (Reschenberg) and evolved to (m + l)-ES (Schwefel) Started with Mutation only (with individual mutation operator) and later added a recombination operator Focus on behavior of individuals August 24, 2000 ECAI 2000 101

Evolutionary Algorithms: EP • Evolutionary Programming (EP) Originally proposed for sequence predictiom and optimal Evolutionary Algorithms: EP • Evolutionary Programming (EP) Originally proposed for sequence predictiom and optimal gaming strategies Currently focused on continuous parameter optimization and training of NNs Could be considered a special case of (m + m) -ES without recombination operator Focus on behavior of species (hence no crossover) Proposed by Larry Fogel (1963) August 24, 2000 ECAI 2000 102

Evolutionary Algorithms: GA • Genetic Algorithms (GA) Perform a randomized search in solution space Evolutionary Algorithms: GA • Genetic Algorithms (GA) Perform a randomized search in solution space using a genotypic rather than a phenotypic Each solution is encoded as a chromosome in a population (a binary, integer, or real-valued string) » Each string’s element represents a particular feature of the solution The string is evaluated by a fitness function to determine the solution’s quality » Better fit solutions survive and produce offspring » Less fit solutions are culled from the population Strings are evolved using mutation & recombination operators. New individuals created by these operators form next generation of solutions Started by Holland (1962; 1975) August 24, 2000 ECAI 2000 103

Evolutionary Algorithms: GP • Genetic Programming (GP) A special case of Genetic Algorithms » Evolutionary Algorithms: GP • Genetic Programming (GP) A special case of Genetic Algorithms » Chromosomes have a hierarchical rather than a linear structure » Their sizes are not predefined » Individuals are tree structured programs » Modified operators are applied to sub trees or single nodes Proposed by Koza (1992) August 24, 2000 ECAI 2000 104

GA Structure (cont. ) • GA Structural Design Selections: Parent Selection Method: » {Proportional GA Structure (cont. ) • GA Structural Design Selections: Parent Selection Method: » {Proportional Roulette, Tournament, Rank, Uniform, . . . } Crossover Operator: » {Once cut, Two cuts, Uniform, BLX, Parent Weighted, . . . } Mutation Operator: » Mutation Rate: {Exponentially Decreasing, Uniform, . . } » Value: {Exponentially Decreasing, Uniform, Normally Distributed, …} August 24, 2000 ECAI 2000 105

GA Parameters (cont. ) • Other possible parameters that could be adjusted: T = GA Parameters (cont. ) • Other possible parameters that could be adjusted: T = Number of Trials = S Ni where N is population size and i= 1, Max_Gen sm = Mutation step s in Normally distributed Mutation value PS = Probability of Selection Parametrized slope of probability distribut. AS = Arity of Parents number of parents in recombination August 24, 2000 ECAI 2000 106

Fuzzy Controller for Mutation Rate: Rule Set GD, PFE-> Pm GD PFE Pm August Fuzzy Controller for Mutation Rate: Rule Set GD, PFE-> Pm GD PFE Pm August 24, 2000 = Genotypic Diversity: Normalized Average Hamming Distance = Percentage Fitness Evaluations: (Completed # Trials) / (Max Allocated # Trials) = Change in Mutation Rate ECAI 2000 107

Statistical Experiments: Set-Up (cont. ) • GA Parameters N Pc Pm G = Base Statistical Experiments: Set-Up (cont. ) • GA Parameters N Pc Pm G = Base Population size: 50 = Crossover rate: 0. 600 = Mutation rate: 0. 005 = Generation Gap 100% replacement Simple GA (SGA) August 24, 2000 25% replacement ECAI 2000 108

Summary of 1200 Experiments J = C*T 2 J = C*e(T-10)/3 Significant change in Summary of 1200 Experiments J = C*T 2 J = C*e(T-10)/3 Significant change in m August 24, 2000 ECAI 2000 Significant change in s 109

Next Steps: Controlling Other Parameters • Run-time Controlled GAs Parameters: DONE Population size: » Next Steps: Controlling Other Parameters • Run-time Controlled GAs Parameters: DONE Population size: » larger size: increase parallel search in solution space » smaller size: focus on current existing regions Probability mutation: » Higher prob. of mutation disrupts current solutions exploration » Lower probability of mutation favors current solutions exploitation • Other Possible Run-time Controllable GAs Parameters: Customized mutation operators: » Variable amount of changes – smaller for good solutions, larger for bad ones Fitness function: » Evolving fitness function (variable weights in multi criteria aggregating function) August 24, 2000 ECAI 2000 110

GAs controlled by FL (cont. ) Probability of Selection: » Parametrized slope distribution ranging GAs controlled by FL (cont. ) Probability of Selection: » Parametrized slope distribution ranging from: – Uniform probability: ignore fitness function and perform random selection of parents extreme case of exploration, to – Proportional selection with rescaling and other intermediate strategies compromise between exploration and exploitation cases, and – Ranking: always select the best N and ignore the rest extreme case of exploitation » Probability as function of fitness and genotypical distance with other solutions enforcing diversity and favoring exploration Probability of crossover: » Constraints applicability to mostly good solutions Customized-crossover operators (for real coded GAs): » Selection of crossovers based on T norms and T conorms causes offsprings to take more extreme values (exploration) » Selection of crossovers based on aggregating operators causes August 24, 2000 111 offsprings to take average. ECAI 2000 (exploitation) values

Fuzzy Controller for DN and DPm: Control Parameters Frequency of Control Actions Control Action: Fuzzy Controller for DN and DPm: Control Parameters Frequency of Control Actions Control Action: mutation rate changed every 10 generations population size change every generation Mutation Rate Mutation rates drops exponentially after a control action that increases it Inference Engine Parameters Left Hand Side (LHS) evaluation: Minimum operator Rule Firing: Rule Output Aggregation: Defuzzification: Minimum operator Maximum operator Center of Gravity (COG) August 24, 2000 ECAI 2000 112

Fuzzy Controller for DN and DPm: Outputs • Outputs N = Change in Population Fuzzy Controller for DN and DPm: Outputs • Outputs N = Change in Population Size (Mult. Factor) D N range is [0. 5, 1. 5] == [Neg High, Pos High] so that. D N range is [0. 5, 1. 5]to 100%High, Pos High] NC corresponds == [Neg of previous Pop Size - so that NC corresponds to 100% of previous Pop Size Population Size is clamped within [25, 150] Pm = Change in Mutation Rate (Mult. Factor) D Pm range is [0. 5, 1. 5] == [Neg High, Pos High] so that NC corresponds to 100% of previous Pm Mutation Rate is clamped within [0. 005, 0. 10] August 24, 2000 ECAI 2000 113

Fusion of Reasoning Models • Develop Collection of Quasi-independent Models • Each Model Generates: Fusion of Reasoning Models • Develop Collection of Quasi-independent Models • Each Model Generates: Output Value (Vi ) - Prediction Confidence parameter (Ci ) derived from training stats. - Introspection • Intelligent Fusion Rules Consider discrepancies among Output values (v) Consider dynamic confidence value (c) associated with each output Example of Fusion for Mortgage Collateral Evaluation Living Area Address (Geo. Coded ) Loc Val e. L e. G AIGEN Lot Size # Beds # Baths, . . . Pool Conditions. . . AICOMP e. C FUSION RULES e. F ei = { V i , C i } August 24, 2000 ECAI 2000 114

Synergy in SC: Reasons & Approaches • SC Leverages Knowledge and Data to Derive Synergy in SC: Reasons & Approaches • SC Leverages Knowledge and Data to Derive the Model • Model = Structure + Parameters (& Search) • Data driven Tuning of Knowledge derived Models Translate domain knowledge to initial structure & parameters Use Global or local data search to tune parameters • Knowledge driven Search Control Use Global or local data search to derive models (Structure + Parameters) Translate domain knowledge into an algorithm’s controller to improve/manage solution convergence and quality • Hybrid Search Methods Embedding local search within global search Embedding knowledge in operators for global search Fusion of models to increase accuracy and reliability August 24, 2000 ECAI 2000 115