1

LEARNING FROM LEARNING FROM DATADATA

Lecture Ten(Chapter 10, Notes;

Chapter 11, Textbook)

2

Chapter 10: Learning From Data

OutlineOutline

The Concept of LearningThe Concept of Learning Data VisualizationData Visualization Neural NetworksNeural Networks

The BasicsThe BasicsSupervised and Unsupervised LearningSupervised and Unsupervised LearningBusiness ApplicationsBusiness Applications

Association RulesAssociation Rules Implications for Knowledge ManagementImplications for Knowledge Management  

3

Chapter 10: Learning From Data

The “Context” of The “Context” of LearningLearning

The “value added” The “value added” collaborative intelligence layer collaborative intelligence layer of of KM architectureKM architecture

Relevant technologies are:Relevant technologies are: Artificial IntelligenceArtificial Intelligence Experts SystemsExperts Systems Case-Based ReasoningCase-Based Reasoning Data WarehousingData Warehousing Intelligent agentsIntelligent agents Neural NetworksNeural Networks

4

Chapter 10: Learning From Data

The “Process” of The “Process” of LearningLearning

A process of filtering and A process of filtering and transforming data into valid transforming data into valid and useful knowledge.and useful knowledge.

Automate via technology Automate via technology tools:tools: provide a collaborative provide a collaborative

learning environment for learning environment for participantsparticipants

enhance their ability to enhance their ability to understand the processes understand the processes / tasks they are dealing / tasks they are dealing withwith

5

Chapter 10: Learning From Data

The “Goals” of LearningThe “Goals” of Learning

Final goal is to Final goal is to improve the improve the qualities of communication qualities of communication and decision makingand decision making

Ways to achieve these Ways to achieve these goals:goals: Verify hypotheses (formed Verify hypotheses (formed

from accumulated from accumulated knowledge)knowledge)

Discover new patterns in Discover new patterns in datadata

Predict future trends and Predict future trends and behaviourbehaviour

6

Chapter 10: Learning From Data

Learning from DataLearning from Data

Build learning models that Build learning models that automatically improve with automatically improve with experience.experience.

Top-down approachTop-down approach Generate ideasGenerate ideas Develop modelsDevelop models Validate modelsValidate models

Bottom-up approachBottom-up approach Discover new (unknown) patternsDiscover new (unknown) patterns Find key relationships in dataFind key relationships in data

7

Chapter 10: Learning From Data

Top-down approach Top-down approach (Example)(Example)

Start with a hypothesis Start with a hypothesis derived from observation or derived from observation or prior knowledgeprior knowledge

““Tourists visiting Egypt Tourists visiting Egypt earn an annual income of at earn an annual income of at least $50,000”least $50,000”

Hypothesis tested by Hypothesis tested by querying database followed querying database followed by analysisby analysis

If tests not supportive, If tests not supportive, hypothesis is revised and hypothesis is revised and test againtest again

8

Chapter 10: Learning From Data

Bottom-up approach Bottom-up approach (Example)(Example)

No hypothesis to testNo hypothesis to test ““Find unknown buying Find unknown buying

patterns by analyzing patterns by analyzing the shopping basket”the shopping basket”

“ … “ … showed married showed married males, age 21 to 27, males, age 21 to 27, shopped for diapers shopped for diapers also brought beer.also brought beer.

““store decided to stack store decided to stack beer cases next to beer cases next to diaper shelf”diaper shelf”

9

Chapter 10: Learning From Data

Data VisualizationData Visualization

Explore visually for trends Explore visually for trends in data useful for making in data useful for making decisiondecision

Exploring data includes:Exploring data includes: Identify key attributes Identify key attributes

and their distributionand their distribution Identify outliersIdentify outliers Extract interesting Extract interesting

grouping of data subsetsgrouping of data subsets Identify initial hypothesisIdentify initial hypothesis

10

Chapter 10: Learning From Data

Example of Data Example of Data VisualizationVisualization

(John Snow and the Cholera outbreak in London, (John Snow and the Cholera outbreak in London, 1845)1845)

11

Chapter 10: Learning From Data

Artificial Neural Artificial Neural Network as Learning Network as Learning

ModelModel Modeled after human Modeled after human

brain’s networkbrain’s network Simulate biological Simulate biological

information information processing via processing via networks of networks of interconnected interconnected neuronsneurons

Neural networks are Neural networks are analog and parallelanalog and parallel

12

Chapter 10: Learning From Data

Neurons – The Basic Neurons – The Basic ElementsElements

The neuron receives The neuron receives inputsinputs, determines , determines their weights their weights (strengths), sums the (strengths), sums the combined inputs, and combined inputs, and compares the total to a compares the total to a thresholdthreshold ((transfer transfer functionfunction))

If total is greater than If total is greater than threshold, the neuron threshold, the neuron fires (sends an fires (sends an outputoutput) )

13

Chapter 10: Learning From Data

A Neuron ModelA Neuron Model

14

Chapter 10: Learning From Data

Learning in Neural Learning in Neural NetworkNetwork

SupervisedSupervised The NN needs a The NN needs a

teacher with a teacher with a training set of training set of examples of input examples of input and outputand output

Unsupervised (or Unsupervised (or Self-Supervised)Self-Supervised) Does not need a Does not need a

teacherteacher

15

Chapter 10: Learning From Data

Supervised LearningSupervised Learning

Each element in a Each element in a training set is paired training set is paired with an acceptable with an acceptable responseresponse

Network makes Network makes successive passes successive passes through the examples through the examples

The weights adjust The weights adjust toward the goal toward the goal state. state.

16

Chapter 10: Learning From Data

A Supervised Neural A Supervised Neural Network (An Example)Network (An Example)

17

Chapter 10: Learning From Data

Unsupervised LearningUnsupervised Learning

No external factors No external factors can influence can influence adjustment of input’s adjustment of input’s weightsweights

No advanced No advanced indication of correct indication of correct or incorrect answersor incorrect answers

Adjusts through Adjusts through direct confrontation direct confrontation with new experienceswith new experiences

18

Chapter 10: Learning From Data

Business Applications Business Applications (1)(1)

Risk managementRisk management

Appraise commercial loan Appraise commercial loan applications applications

NN trained on thousands of NN trained on thousands of applications, half of which were applications, half of which were approved and the other half rejected approved and the other half rejected by the bank’s loan officersby the bank’s loan officers

Through supervised learning, NN Through supervised learning, NN learned to pick risks that constitute a learned to pick risks that constitute a bad loanbad loan

Identifies loan applicants who are Identifies loan applicants who are likely to default on their paymentslikely to default on their payments

19

Chapter 10: Learning From Data

Business Applications (2)Business Applications (2)

Predicting Foreign Exchange Predicting Foreign Exchange Fluctuations:Fluctuations: A set of relevant indicators were A set of relevant indicators were

identified, used as inputs to NNidentified, used as inputs to NN NN was trained for exchange rates NN was trained for exchange rates

of US dollar against Swiss franc of US dollar against Swiss franc and Japanese yen, using data from and Japanese yen, using data from first 6 months of 1990. Then it was first 6 months of 1990. Then it was tested over an 8- to 11-week periodtested over an 8- to 11-week period

Results revealed return on capital Results revealed return on capital of about 20% of about 20%

20

Chapter 10: Learning From Data

Business Applications (3)Business Applications (3)

Mortgage Appraisals:Mortgage Appraisals: Neural network uses the data in Neural network uses the data in

the mortgage loan application the mortgage loan application It estimates value of the property It estimates value of the property

based on the immediate based on the immediate neighborhood, the city, and the neighborhood, the city, and the countrycountry

The system comes up with a The system comes up with a valuation for the property and a valuation for the property and a risk analysis for the loan. risk analysis for the loan.

21

Chapter 10: Learning From Data

Association RulesAssociation Rules

A KB tool that generates a A KB tool that generates a set of rules to help set of rules to help understanding understanding relationships that exist in relationships that exist in datadata

Types:Types: Boolean ruleBoolean rule Quantitative ruleQuantitative rule Multi-dimensional ruleMulti-dimensional rule Multi-level association Multi-level association

rulerule

22

Chapter 10: Learning From Data

Boolean Rule (An Boolean Rule (An Example)Example)

A rule that examines the A rule that examines the presence or absence of presence or absence of itemsitems

For example, if a For example, if a customer buys a PC and a customer buys a PC and a 17” monitor, then he will 17” monitor, then he will buy a printer. Presence buy a printer. Presence of items (a PC and 17” of items (a PC and 17” monitor) implies presence monitor) implies presence of the printer in the of the printer in the customer’s buying listcustomer’s buying list

23

Chapter 10: Learning From Data

Quantitative Rule (An Quantitative Rule (An Example)Example)

A rule that considers the A rule that considers the quantitative values of quantitative values of itemsitems

For example, if a For example, if a customer earns between customer earns between $30,000 and $50,000 $30,000 and $50,000 and owns an apartment and owns an apartment worth between $250,000 worth between $250,000 and $500,000, he will and $500,000, he will buy a 4-door automobilebuy a 4-door automobile

24

Chapter 10: Learning From Data

Multi-dimensional RuleMulti-dimensional Rule

A rule that refers to a A rule that refers to a multitude of dimensionsmultitude of dimensions

If a customer lives in a If a customer lives in a big city and earns more big city and earns more than $35,000, then he than $35,000, then he will buy a cellular phonewill buy a cellular phone

This rule involves 3 This rule involves 3 attributes: attributes: living, living, earning, and buyingearning, and buying. . Therefore, it is a multi-Therefore, it is a multi-dimensional ruledimensional rule

25

Chapter 10: Learning From Data

Multi-level Association Multi-level Association RuleRule

26

Chapter 10: Learning From Data

Implications for Implications for Knowledge Knowledge

ManagementManagement Cost / benefit analysisCost / benefit analysis

Tangible costs - user training, Tangible costs - user training, hardware + software, backup, hardware + software, backup, support, maintenancesupport, maintenance

Intangible costs - user resistance Intangible costs - user resistance and learning curveand learning curve

Quality AssuranceQuality Assurance Adequacy of initial designAdequacy of initial design Level and frequency of Level and frequency of

maintenancemaintenance

27

Chapter 10: Learning From Data

User Interface(Web browser software installed on each user’s PC)

Authorized access control(e.g., security, passwords, firewalls, authentication)

Collaborative intelligence and filtering(intelligent agents, network mining, customization, personalization)

Knowledge-enabling applications(customized applications, skills directories, videoconferencing, decision support systems,

group decision support systems tools)

Transport(e-mail, Internet/Web site, TCP/IP protocol to manage traffic flow)

Middleware(specialized software for network management, security, etc.)

The Physical Layer(repositories, cables)

. . . . .

Databases Data warehousing(data cleansing,

data mining)

Groupware(document exchange,

collaboration)

Legacy applications(e.g., payroll)

1

2

3

4

5

6

7

Layers of KM Layers of KM ArchitectureArchitecture