1 CHAPTER 16 Neural Computing Applications, and Advanced Artificial Intelligent Systems and Applications

1

CHAPTER 16

Neural Computing Applications, and Advanced Artificial Intelligent

Systems and Applications

2

Neural Computing Applications, and Advanced Artificial Intelligent

Systems and Applications

Several Real-World Applications of ANN Technology Advanced AI Systems

– Genetic Algorithms

– Fuzzy Logic

– Qualitative Reasoning

Integration (Hybrids)

Decision Support Systems and Intelligent Systems, Efraim Turban and Jay E. Aronson6th ed, Copyright 2001, Prentice Hall, Upper Saddle River, NJ

3

Areas of ANN Applications:An Overview

Representative Business ANN Applications

Accounting Finance Human Resources Management Marketing Operations


4

Accounting

Identify tax fraud Enhance auditing by finding irregularities


5

Signatures and bank note verifications Mortgage underwriting Foreign exchange rate forecasting Country risk rating Bankruptcy prediction Customer credit scoring Credit card approval and fraud detection Stock and commodity selection and trading

Finance


6

Credit card profitability Forecasting economic turning points Bond rating and trading Pricing initial public offerings Loan approvals Economic and financial forecasting Risk management

Finance 2


7

Predicting employees’ performance and behavior Determining personnel resource requirements

Human Resources


8

Management

Corporate merger prediction Country risk rating


9

Consumer spending pattern classification Customers’ characteristics Sales forecasts Data mining Airline fare management Direct mail optimization Targeted marketing

Marketing


10

Operations Airline crew scheduling Predicting airline seat demand Vehicle routing Assembly and packaged goods inspection Quality control Matching jobs to candidates Production/job scheduling Factory process control

Many More


11

Credit Approval with Neural Networks

Increases loan processor productivity by 25 to 35% over other computerized tools

Also detects credit card fraud


12

The ANN Method

Data from the application and into a database

Preprocess applications manually

Neural network trained in advance with many good and bad risk cases


13

Neural Network Credit AuthorizerConstruction Process

Step 1: Collect data

Step 2: Separate data into training and test sets

Step 3: Transform data into network inputs

Step 4: Select, train, and test network

Step 5: Deploy developed network application


14

Bankruptcy Prediction with Neural Networks

Concept Phase

Paradigm: Three-layer network, back-propagation

Training data: Small set of well-known financial ratios

Data available on bankruptcy outcomes

Supervised network

Training time not to be a problem


15

Application Design

Five Input Nodes

X1: Working capital/total assets

X2: Retained earnings/total assets

X3: Earnings before interest and taxes/total assets

X4: Market value of equity/total debt

X5: Sales/total assets

Single Output Node: Final classification for each firm – Bankruptcy or – Nonbankruptcy

Development Tool: NeuroShell


16

Development– Three-layer network with backpropagation (Figure 16.3)

– Continuous valued input

– Single output node: 0 = bankrupt, 1 = not bankrupt

Training– Data Set: 129 firms

– Training Set: 74 firms; 38 bankrupt, 36 not

– Ratios computed and stored in input files for:• The neural network

• A conventional discriminant analysis program


17

Architecture of the Bankruptcy Prediction Neural Network

(Figure 16.3)

X4

X3

X5

X1

X2Bankrupt 0

Not bankrupt 1


18

Parameters– Learning threshold– Learning rate– Momentum

Testing– Two Ways

• Test data set: 27 bankrupt firms, 28 nonbankrupt firms• Comparison with discriminant analysis

– The neural network correctly predicted:• 81.5 percent bankrupt cases • 82.1 percent nonbankrupt cases


19

ANN did better predicting 22 out of the 27 actual cases

Discriminant analysis predicted only 16 correctly

Error Analysis– Five bankrupt firms misclassified by both methods

– Similar for nonbankrupt firms

Neural network at least as good as conventional

Accuracy of about 80 percent is usually acceptable for neural network applications


20

Stock Market Prediction System with Modular Neural Networks

Accurate Stock Market Prediction - Complex Problem

Several Mathematical Models - Disappointing Results

Fujitsu and Nikko Securities: TOPIX Buying and Selling Prediction System


21

Input: Several technical and economic indexes

Several modular neural networks relate past indexes, and buy/sell timing

Prediction system– Modular neural networks

– Very accurate


22

Network Architecture(Figure 16.4)

Network Model: 3 layers, standard sigmoid function, continuous output [0, 1]

High-speed Supplementary Learning Algorithm

Training Data – Data Selection

– Training Data


23

Preprocessing: Input Indexes - Converted into spatial patterns, preprocessed to regularize them

Moving Simulation Prediction Method (Figure 16.5)

Result of Simulations– Simulation for Buying and Selling Stocks

– Example (Figure 16.6)

– Excellent Profit


24

Integrated ANNs and Expert Systems

1. Resource Requirements Advisor

– Advises users on database systems’ resource requirements

– Predicts the time and effort to finish a database project

– ES shell AUBREY and neural network tool NeuroShell

– ES supported data collection

– ANN used for data evaluation

– ES final analysis


25

2. Personnel Resource Requirements Advisor

– Project personnel resource requirements for maintaining networks or workstations at NASA

– Rule-based ES determines the final resource projections

– ANN provides project completion times for services requested(Figure 16.7)


26

3. Diagnostic System for an Airline

– Singapore Airlines

– Assists technicians in diagnosing avionics equipment

– INSIDE (Inertial Navigation System Interactive Diagnostic Expert)

– Designed to reduce the diagnostic time(Figure 16.8)


27

4. Manufacturing Product Liability

– United Technologies Carrier

– Two ES + ANN

– Patterns fed into multilayer feedforward ANN

– Integrated with a database into an Automatic Early Warning System (AEWS)


28

5. Oil Refinery Production Scheduling and Environmental Control– Citgo Petroleum Corporation

– Lower costs

– Improved safety

– Higher product quality

– Higher yields


29

Genetic Algorithms

Goal (evolutionary algorithms): Demonstrate self-organization and adaptation by exposure to the environment

System learns to adapt to changes. Example 1: Vector Game

– Random trial and error

– Genetic algorithm solution

Process (Figure 16.9) Example: the game of MasterMind


30

Genetic Algorithm

Definition and Process Genetic algorithm: "an iterative procedure maintaining a

population of structures that are candidate solutions to specific domain challenges” (Grefenstette [1982])

Each candidate solution is called a chromosome

Chromosomes can copy themselves, mate, and mutate

Use specific genetic operators - reproduction, crossover and mutation


31

Primary Operators of Most Genetic Algorithms

Reproduction

Crossover

Mutation


32

Genetic Algorithm Operators

1 0 1 0 1 1 1

1 1 0 0 0 1 1

Parent 1

Parent 2

1 0 1 0 0 1 1

1 1 0 0 1 1 0

Child 1

Child 2 Mutation


33

GA Example: The Knapsack Problem

Item: 1 2 3 4 5 6 7 Benefit: 5 8 3 2 7 9 4 Weight: 7 8 4 10 4 6 4 Knapsack holds a maximum of 22 pounds Fill it to get the maximum benefit Solutions take the form of a string of 1’s Solution: 1 1 0 0 1 0 0 Means choose items 1, 2, 5. Weight = 21, Benefit = 20 Evolver solution in Figure 16.10


34

Genetic AlgorithmsApplications and Software

Type of machine learning

Set of efficient, domain-independent search heuristics for a broad spectrum of applications


35

Genetic Algorithm Application Areas

Dynamic process control Induction of rule optimization Discovering new connectivity topologies Simulating biological models of behavior and evolution Complex design of engineering structures Pattern recognition Scheduling Transportation Layout and circuit design Telecommunication Graph-based problems


36

Business Applications

Channel 4 Television (England) to schedule commercials Driver scheduling in a public transportation system Jobshop scheduling Assignment of destinations to sources Trading stocks Productivity in whisky-making is increased

Often genetic algorithm hybrids with other AI methods


37

Representative Commercial Packages

Evolver (Excel spreadsheet add-in) Genetic Algorithm User Interface (GAUI) OOGA (Object-Oriented GA for industrial use) XperRule Genasys (ES shell with an embedded genetic

algorithm) Sugal Genetic Algorithm Simulator


38

Optimization Algorithms

Via neural computing sometimes

Genetic algorithms and their derivatives can optimize (or nearly optimize) complex problems


39

Fuzzy Logic

Fuzzy logic deals with uncertainty

Uses the mathematical theory of fuzzy sets

Simulates the process of normal human reasoning

Allows the computer to behave less precisely and logically

Decision making involves gray areas and the term maybe


40

Fuzzy Logic Advantages

Provides flexibility Provides options Frees the imagination More forgiving Allows for observation Shortens system development time Increases the system's maintainability Uses less expensive hardware Handles control or decision-making problems not

easily defined by mathematical models


41

Fuzzy Logic Example:What is Tall?

In-Class ExerciseProportion

Height Voted for5’10” 0.055’11” 0.106’ 0.606’1” 0.156’2” 0.10

– Jack is 6 feet tall– Probability theory - cumulative probability – There is a 75 percent chance that Jack is tall


42

Fuzzy logic - Jack's degree of membership within the set of tall people is 0.75

We are not completely sure whether he is tall or not Fuzzy logic - We agree that Jack is more or less tall Membership Function

< Jack, 0.75 Tall >

Knowledge-based system approach: Jack is tall (CF = .75)

Belief functions

Can use fuzzy logic in rule-based systems


43

Membership Functions in Fuzzy Sets (Figure 16.11)


Membership

Short Medium Tall

Height in inches (1 inch = 2.54 cm)

0.5

1.0

64 69 74

44

Fuzzy Logic Applications and

Software Difficult to apply when people provide evidence

Used in consumer products that have sensors– Air conditioners

– Cameras

– Dishwashers

– Microwaves

– Toasters

Special software packages

Controls applications


45

Examples of Fuzzy Logic

Example 1: Strategic planning– STRATASSIST - fuzzy expert system that helps small- to

medium-sized firms plan strategically for a single product

Example 2: Fuzziness in real estate

Example 3: A fuzzy bond evaluation system


46

Fuzzy Logic Software

Fuzzy Inference Development Environment (FIDE)

Z Search HyperLogic Corporation demos Others


47

Qualitative Reasoning (QR)

– Means of representing and making inferences using general, physical knowledge about the world

– QR is a model-based procedure that consequently incorporates deep knowledge about a problem domain

– Typical QR Logic• “If you touch a kettle full of boiling water on a stove, you

will burn yourself”

• “If you throw an object off a building, it will go down”


48

But

No specific knowledge about boiling temperature, just that it is really hot!

No specific information about the building or object, unless you are the object, or you are trying to catch it


49

Main goal of QR: To represent common sense knowledge about the physical world, and the underlying abstractions used in quantitative models (objects fall)

Given such knowledge and appropriate reasoning methods, an ES could make predictions and diagnoses, and explain the behavior of physical systems qualitatively, even when exact quantitative descriptions are unavailable or intractable


50

Relevant behavior is modeled

Temporal and spatial qualities in decision making are represented effectively

Applies common sense mathematical rules to variables and functions

There are structure rules and behavior rules

Qualitative Reasoning


51

Some Real-World QR Applications

Nuclear plant fault diagnoses

Business processes

Financial markets

Economic systems


52

Intelligent Systems Integration

Combine – Neural Computing

– Expert Systems

– Genetic Algorithms

– Fuzzy Logic

Example: International investment management--stock selection

Fuzzy Logic and ANN (FuzzyNet) to forecast the expected returns from stocks, cash, bonds, and other assets to determine the optimal allocation of assets


53

Global markets Integrated network architecture of the system

(Figure 16.12)

Technologies Expert system (rule-based) for country and stock selection Neural network for forecasting Fuzzy logic for assessing factors without reliable data


54

FuzzyNet Architecture

(Figure 16.12)

Membership Function Generator (MFG)

Fuzzy Information Processor (FIP)

Back-propagation Neural Network (BPN)


55

Data Mining and KnowledgeDiscovery in Databases (KDD)

Hidden value in data Knowledge Discovery in Databases (KDD)


56

The KDD ProcessStart with Raw Data and Do

1. Selection to produce target the appropriate data which undergoes

2. Preprocessing to filter the data in preparation for

3. Transformation so that

4. Data Mining can identify patterns that go through

5. Interpretation and Evaluation resulting in knowledge


57

Data Mining

Find kernels of value in raw data ore

Theoretical advances

– Knowledge discovery in textual databases

– Methods based on statistics, cluster analysis, discriminant analysis, fuzzy logic, genetic algorithms, and neural networks

– Ideal for data mining


58

AI Methods and Data Mining for Search

Neural Networks

Expert Systems

Rule Induction


59

Data Mining Applications Areas

Marketing

Investment

Fraud detection

Manufacturing


60

Information Overload

Data mining methods can sift through soft information to identify relationships automatically

Intelligent agents


61

Important KDD andData Mining Challenges

Dealing with larger databases

Working with higher dimensionalities of data

Overfitting--modeling noise rather than data patterns

Assessing statistical significance of results

Working with constantly changing data and knowledge

Continue


62

Working through missing and noisy data

Determining complex relationships between fields

Making patterns more understandable to humans

Providing better user interaction and prior knowledge about the data

Providing integration with other systems


Documents

1 CHAPTER 16 Neural Computing Applications, and Advanced Artificial Intelligent Systems and Applications