LEARNING FROM DATA

cracklegulleyΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 4 χρόνια και 20 μέρες)

45 εμφανίσεις

1

LEARNING FROM DATA

Lecture Ten

(Chapter 10, Notes;

Chapter 11, Textbook)


Chapter 10: Learning From Data

2

Outline


The Concept of Learning


Data Visualization


Neural Networks



The Basics



Supervised and Unsupervised Learning



Business Applications


Association Rules


Implications for Knowledge Management





Chapter 10: Learning From Data

3

The

Context


of Learning


The

value added


collaborative intelligence
layer
of KM architecture


Relevant technologies are:


Artificial Intelligence


Experts Systems


Case
-
Based Reasoning


Data Warehousing


Intelligent agents


Neural Networks


Chapter 10: Learning From Data

4

The

Process


of Learning


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


Automate via technology tools:


provide a collaborative
learning environment for
participants


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


Chapter 10: Learning From Data

5

The

Goals


of Learning


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


Ways to achieve these goals:


Verify hypotheses (formed
from accumulated knowledge)


Discover new patterns in data


Predict future trends and
behaviour


Chapter 10: Learning From Data

6

Learning from Data


Build learning models that automatically
improve with experience.


Top
-
down approach


Generate ideas


Develop models


Validate models


Bottom
-
up approach


Discover new (unknown) patterns


Find key relationships in data


Chapter 10: Learning From Data

7

Top
-
down approach (Example)


Start with a hypothesis derived
from observation or prior
knowledge



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



Hypothesis tested by querying
database followed by analysis


If tests not supportive, hypothesis
is revised and test again



Chapter 10: Learning From Data

8

Bottom
-
up approach (Example)


No hypothesis to test



Find unknown buying
patterns by analyzing the
shopping basket







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



store decided to stack beer
cases next to diaper shelf



Chapter 10: Learning From Data

9

Data Visualization


Explore visually for trends in
data useful for making decision


Exploring data includes:


Identify key attributes and
their distribution


Identify outliers


Extract interesting grouping
of data subsets


Identify initial hypothesis


Chapter 10: Learning From Data

10

Example of Data Visualization

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


Chapter 10: Learning From Data

11

Artificial Neural Network as
Learning Model


Modeled after human brain’s
network


Simulate biological
information processing via
networks of interconnected
neurons



Neural networks are analog
and parallel


Chapter 10: Learning From Data

12

Neurons


The Basic Elements


The neuron receives
inputs
, determines their
weights (strengths), sums
the combined inputs, and
compares the total to a
threshold

(
transfer
function
)


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


Chapter 10: Learning From Data

13

A Neuron Model


Chapter 10: Learning From Data

14

Learning in Neural Network


Supervised


The NN needs a teacher
with a training set of
examples of input and
output


Unsupervised (or Self
-
Supervised)


Does not need a teacher


Chapter 10: Learning From Data

15

Supervised Learning


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


Network makes successive
passes through the
examples


The weights adjust toward
the goal state.


Chapter 10: Learning From Data

16

A Supervised Neural Network
(An Example)


Chapter 10: Learning From Data

17

Unsupervised Learning


No external factors can
influence adjustment of
input

s weights


No advanced indication
of correct or incorrect
answers


Adjusts through direct
confrontation with new
experiences



Chapter 10: Learning From Data

18

Business Applications (1)


Risk management



Appraise commercial loan applications


NN trained on thousands of applications,
half of which were approved and the other
half rejected by the bank

s loan officers


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


Identifies loan applicants who are likely to
default on their payments



Chapter 10: Learning From Data

19

Business Applications (2)


Predicting Foreign Exchange Fluctuations:


A set of relevant indicators were identified,
used as inputs to NN


NN was trained for exchange rates of US
dollar against Swiss franc and Japanese yen,
using data from first 6 months of 1990.
Then it was tested over an 8
-

to 11
-
week
period


Results revealed return on capital of about
20%


Chapter 10: Learning From Data

20

Business Applications (3)


Mortgage Appraisals:


Neural network uses the data in the
mortgage loan application


It estimates value of the property based
on the immediate neighborhood, the
city, and the country


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


Chapter 10: Learning From Data

21

Association Rules


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


Types:


Boolean rule


Quantitative rule


Multi
-
dimensional rule


Multi
-
level association rule


Chapter 10: Learning From Data

22

Boolean Rule (An Example)


A rule that examines the
presence or absence of items


For example, if a customer
buys a PC and a 17


monitor,
then he will buy a printer.
Presence of items (a PC and
17


monitor) implies
presence of the printer in the
customer

s buying list


Chapter 10: Learning From Data

23

Quantitative Rule (An
Example)


A rule that considers the
quantitative values of items


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


Chapter 10: Learning From Data

24

Multi
-
dimensional Rule


A rule that refers to a
multitude of dimensions


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


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



Chapter 10: Learning From Data

25

Multi
-
level Association Rule



Chapter 10: Learning From Data

26

Implications for Knowledge
Management


Cost / benefit analysis


Tangible costs
-

user training,
hardware + software, backup,
support, maintenance


Intangible costs
-

user resistance and
learning curve


Quality Assurance


Adequacy of initial design


Level and frequency of maintenance


Chapter 10: Learning From Data

27

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 Architecture