Closing Some Loose Ends

yawnknotΔιαχείριση

6 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

63 εμφανίσεις

Closing Some Loose Ends

Sources:



David W. Aha


My
own


Thomas
H. Davenport, Laurence
Prusak
, 1998

Classifiers

Classifiers



Case
-
based reasoning (CBR) classifier





Induction of decision trees (IDT)





CBR+IDT classifier




Others (e.g., covered in the Data Mining course):


Support Vector Machines


Linear regression


Neural networks






So which one is best?



No Free Lunch Theorem



Each of these classifiers have a
bias




To explain the bias, let us examine a situation where
instances (or cases) are pairs of numeric features and a binary
classification problem:




((
x,y
),class)




Let us draw the space: CBR, K
-
d trees (K=2), Support vector
machines




Let us construct examples where each of these classifiers
works best



How does the other classifiers work on these
examples?



Formulation of the no free lunch theorem


Knowledge Management

The Beginning: The Apollo 13
Situation

http://www.youtube.com/watch?v=nEl0NsYn1fU


The oxygen tanks had originally been designed to run off the
28 volt DC















The tanks were redesigned to also run off the 65 volt DC


The Changing Game

The New Economics


Manufacturing


Service


Tangible


Intangible


Consumable


Inconsumable


Structural


Intellectual



Tobin’s Q ratio


company’s stock market value / value of its physical assets



Is increasing dramatically. What does this mean?

Increasing importance of intellectual capital in the
United States (Barr & Magaldi, 1996)


Knowledge Management (KM)

An increasingly important new business movement that
promotes the creation, sharing, & leveraging of knowledge
within an organization to maximize business results.


Effective tools to capture,


leverage & reuse knowledge

Technology

Develop a culture

for knowledge sharing

Organizational Dynamics

Needs

Financial constraints

Loss of organizational knowledge

Needs

Problems:

Knowledge Management: Issues



Technical and Business Expertise:


Proficiencies


Know
-
How


Skills



Work Practice Execution:


Processes


Methodologies


Practices


Lessons learned


Why Knowledge Management?


Leverages Core Business Competence



Accelerates Innovation (Time to Market)



Improves Cycle Times (Market to Collection)



Improves Decision Making



Strengthens Organizational Commitment



Builds sustainable differentiation



CBR: The Knowledge Management Plunge

“Case
-
based reasoning programs have been shown to bring
about marked improvements in customer service.”


-

Thomas H. Davenport, Laurence Prusak, 1998


-

Working Knowledge: How Organizations Manage What They Know

KM

CBRWorks

eGain eService Enterprise (E3)


KM Project Domains: CBR Applicable?

(
KM World
, 1/99, Dan Holtshouse, Xerox)

1. Sharing knowledge and best practices

2. Instilling responsibility for knowledge sharing

3. Capturing and reusing past experiences

4. Embedding knowledge
(products/services/processes)

5. Producing knowledge as a product

6. Driving knowledge generation for innovation

7. Mapping networks of experts

8. Building/mining customer knowledge bases

9. Understanding/mining customer knowledge bases

10. Leveraging intellectual assets.

KM Domains/Tasks

CBR Applicable?

Yes

No

Yes

Yes


Yes


No

Yes

No

Yes


No

1999 AAAI KM/CBR Workshop

~45 attendees:
Siemens, Schlumberger, Motorola, NEC, British
Airways, General Motors, Boeing, Ford Motor Company, World Bank

Goals:

1. Explain KM issues to CBR researchers

2. Report on recent CBR approaches for KM tasks

3. Share cautions, knowledge, & experiences

Some observations
:

1. Embedded/integrated in
knowledge

processes

2. Benefits of semi
-
structured case representations

3. Interactive (“conversational”) systems

Limitations of CBR for KM

(from the 1999 AAAI KM/CBR Workshop)

1. Main limitation is time and effort? (Wess/Haley)

2. Limitations from working with simple representations (Haley)


Becoming less problematic (e.g., with development of textual CBR)

3. Rule
-
based integrations


Suffer from old problems of rule acquisition


But KM problem
-
solving techniques are combating this (Studer)

4. More intuitive case authoring capabilities

5. Tools for working with heterogeneous data sources

Panel: Lessons & Suggested Directions

CBR Roles:


Accumulate, extend, preserve, distribute, reuse corporate knowledge


Extracting tacit knowledge


Customer relationship management

Lessons & Observations:


Integrate CBR with KM tasks & task models


Integrate case retrieval with presentation with tools/workplaces


Integrate case construction/indexing with work product development


Need more advanced (automated) case authoring tools


Must consider effects on user groups, time, organizational impact


CBR not a complete KM solution

Experience Management vs CBR

Experience
Management

CBR

(Organization)

(IDSS)

2.
Reuse

3.
Revise

4.
Retain

Case
Library

1. Retrieve

Background
Knowledge

Experience
base

Reuse
-
related
knowledge

Problem
acquisition

Experience
evaluation
and retrieval

Experience
adaptation

Experience
presentation

Complex
problem
solving

Development and
Management
Methodologies

BOOK

Relating KM with AI

AI

Knowledge
-
Based

Systems

Human

Factors

KM

Business

Processing

CBR

Distinguishing KM from Data Mining

KDD Focus:


Large databases


Autonomous pattern recognition

Knowledge Discovery from Databases Process:

Database Acquisition

Data Warehousing

Data Cleansing

Data Mining

Data Maintenance

KM Focus:


Capturing organizational dynamics processes


Interaction (i.e., decision support)


Process
-
Oriented CBR


Most KM tasks are performed in the context of a well
-
defined (e.g., business) process, and any techniques
designed to support KM must be embedded in this process

KM examples (many):



Enterprise resource planning (O’Leary)



Project process (Maurer & Holz)

CBR examples (few):


Leake et al.: Feasibility assessment in design process


Moussavi, Shimazu: Cases represent processes


Reddy & Munoz
-
Avila: Project Planning

Motivation for Design Project


Embedding CBR into existing tools has been shown to be
an effective way to insert CBR into KM processes



We saw it this year in a number of projects:


Help
-
desk for LTS


Recommender for university events


Companies processes



We discuss two applications


They have a similar flavor to most of the design
projects

Two Examples

EXTERNAL MONITORING

Alerts

Spiders

Workflow

Scheduling

Collaboration

Suspenses

Records
Management

Document
Management

E
-
mail

OA
tools

Library catalog

Online
databases

E
-
journals

How
-
to guides

Document
Delivery Service

Bulletin
boards

Buckets

Profiles

MIS

INFORMATION SOURCES

WORKSPACE

PERSONAL
PORTAL

AFRL Proposed KM Environment

(multi?) impersonal

Personalization

Semantic Web


Ontologies

DS1

DS2

DS3

Distributed

data sources

Assistant

Agent

Case Repository


Causal Model

Current Problem

User Ontologies

Personal Portal/

Workspace

Information

Sources

Individualized Portal

Information
Domains

Data
Systems

Virtual
Library

Buckets

Finance

Personnel

A
B
C
D
Executive Information System

Out
-
of
-
Family Disposition (OOFD) Process

NASA
-
Kennedy Space Center:

Shuttle Processing Directorate

CBR expertise

Topic:
Performing project tasks outside range of expertise



Lack of task familiarity

Motivations:
Downsizing, employee loss, technology pace

Resources:
Interim problem reports



Standardized text documents for reporting problems/solutions



Given: 12 of these reports

Pre
-
flight, launch,

landing, recovery

Prof. I. Becerra
-
Fernandez

Has this data already
been gathered? If so,
WHERE?
NO, need
to gather data
NO, need
new mission
Science
Data
Need
YES, here is the DATA!
YES, recommend this
OBSERVATORY!
Science Mission Parameters
Science Mission Assistant and Research Tool (SMART)
Intelligent Data
Prospector (IDP)
Intelligent Resource
Prospector (IRP)
Design Assistant
(IMDA)
Intelligent Mission
Can an
existing resource
obtain the data for me?
If so,
WHAT?
I would like to
formulate
a new mission…
HOW?

Example KM Aplication: SMART KM
Portal

SMART:
S
cience
M
ission
A
ssistant &
R
esearch
T
ool

Categorization
: An interactive, web
-
based tool suite

Purpose
:
Reduce time/cost required to define new science initiatives

Uncertainty

SMART is Architected as a Web Portal

SMART
User

Web

Browser

http://smart.gsfc.nasa.gov

SMART

Intelligent Data Prospector


Find data sets

Intelligent Resource Prospector


Find an observatory

Intelligent Mission Design Asst


Design a science mission

http://smart.gsfc.nasa.gov/irp/

Browse Observatory Knowledge Base

Map


Tree


Observatory Lists

Search Observatory Knowledge Base


Word/Phrase Search


Interactive Dialog

Discussions

Experts

SMART

Intelligent Resource Prospector

http://smart.gsfc.nasa.gov/imda/

Browse Mission Knowledge Base

Map


Tree


Mission Lists

Search Mission Knowledge Base


Word/Phrase Search


Interactive Dialog

Discussions

Experts

Design a Mission

SMART

Intelligent Mission Design Asst

SMART

Concept Map Viewer:
Observatories

SMART

Hierarchical Directory

Viewer

SMART

Database Views

SMART

Conversational CBR

Question/Response

Interface

SMART

Collaborative

Discussions Interface

SMART IMDA

Design a Mission

Create/Edit a Mission

Validate Design


Power Design Advisor


Thermal Design Advisor


Communications



Design Advisor





Invoke

Design

Validation

Agent

(applet)

(server

DB

access)

(applet)

(KM tool

service)

(KM tool

service)

(expert

systems)

Searching for Missions Using CCBR

SMART

Conversational Mission Search Engine

Describe what you are looking for:

“I’m looking for astronomy missions in low
-
Earth orbit.”


Ranked questions:

Score Answer Name Title


“X
-
ray”

Q17

What portion of the spectrum is observed?

60


Q7

What launch vehicle?

50


Q32

What mission phase?

20


Q23

Low or high inclination orbit?

10


Q41

Cryogenically
-
cooled instrument?

Ranked cases:

Score

Name

Title

90

XTE

X
-
Ray Timing Explorer

90

AXAF

Chandra X
-
Ray Observatory

30

GRO

Gamma Ray Observatory

30

EUVE

Extreme Ultra
-
Violet Explorer


Question:
Q17

Title:

What portion of the spectrum is
observed?

Description:
What portion of the
electro
-
magnetic spectrum are you
interested in?

Select your answer:




Visible light




Infra
-
red




Ultra
-
violet





Microwave




X
-
Ray





Radiowave




Gamma Ray




Lessons Learned

Keywords
: Philippines, evacuation, disaster relief, c
2
, NEO, Fiery Vigil, etc.

Observation
: Assignment of air traffic controllers to augment host country
controllers was critical to safe evacuation airfield operation.

Discussion
: The rapid build
-
up of military flight operations…overloaded the
civilian host nation controllers. Military controllers maintained 24 hour
operations. ...

Lesson Learned
: Military air traffic controllers are required whenever a civilian
airport is transformed into an intensive military operating area for contingency
operations.

Recommended Action
: Ensure controllers and liaison teams are part of the
evacuation package, and establish early liaison with host nation to coordinate an
agreement on operational procedures.

What

How

When

Joint Unified Lessons Learned System (JULLS)

Database
: 908 “scrubbed” lessons from the CINC’s (1991
-
)


Unclassified subset: 150 lessons (Armed Forces Staff College)


33 relate to NEOs

Lesson Format
: 43 attributes


e.g., ID Number, submitting command, subject, date


Unified Joint Task List number


Content attributes: All in text format


Keywords


Observation


Discussion


Lesson

learned


Recommended action

Some Lessons Learned Centers/Systems

Air Force


o Air Force Automated Lessons Learned Capture and Retrieval System


o Air Force Center for Knowledge Sharing Lessons Learned


o Air Combat Command Center for Lessons Learned


o Automated Lessons Learned Collection & Retrieval System

Army


o Center for Army Lessons Learned (CALL)


o SARDA: Contracting Lessons Learned


o US Army Europe
-

Lessons Learned System

Coast Guard


o Coast Guard Universal Lessons Learned

Joint Forces


o JCLL: Joint Center for Lessons Learned

Marine Corps


o Marine Corps Lessons Learned System

Navy


o NDC: Navy Doctrine Command Lessons Learned System


o NAWCAD: Navy Combined Automated Lessons Learned


o NAVFAC: Naval Facilities Engineering Command Lessons Learned System


Government (non
-
military)


o NASA Lessons Learned Information System


o International Safety Lessons Learned Information System


o NASA
-
Goddard: RECALL: Reusable Experience with CBR for Automating Lessons Learned)


o NIST: Best Practices Hyperlinks


o DoE: US Department of Energy Lessons Learned


Other


o Canadian Army Lessons Learned Centre


o United Nations: UN Lessons Learned in Peacekeeping Operations

Lessons Learned Repositories: Functionality


Center for

Lessons


Learned

Documented Lessons

Decision
-
Support

Tool

Retrieval

Tool

Interface

Lessons

Learned

Repository

Lessons Learned System

Search queries

Relevant

lessons

Lessons Learned Systems:

Unrealistic Assumptions

The decision maker

1. has
time

to search for lessons,

2. knows
where

to search for lessons,

3. knows
how

to search for lessons, and

4. knows
how to interpret

retrieved lessons for their
current decision
-
making context.

Decision Support

Tool












User

Interface

Active

Lessons Learned Repositories

Center for

Lessons


Learned

Documented

Lessons

Retrieval

Tool

Interface

Lessons

Learned

Repository

Lessons Learned System

LL Agent
: (CBR)


Relevance
Assessment


Retrieval


Interpretation

Search queries

Relevant

lessons

Issues for Active Lessons Learned

Documented Lessons

LL Agent

(CBR)

User

Case Library

Case extraction

Decision Support

Tool

Decision
-
Making Process

1. Case extraction methods

2. Case representation

3. Choice of decision support tool

4. Embedded LL agent behavior

Lessons Learned: NEO Critiquing Example

Compose an

Intermediate


Stage Base

Tasks

Scenario
:


50 miles from ISB #1


30 miles from ISB #2


Commercial airfield

Resources
:


Transport vehicles





Joint Air Command


Military air traffic controller


...

Objects
:

1. Planning tasks

2. Resources

3. Assignments

4. Task relations

5. Scenario


Coordinate

with local

security forces

Coordinate with

airfield traffic controllers

...

Lesson Learned #13167
-
92740
:


Index
: Coordinate w/ traffic controllers


Lesson
: If ISB is a commercial airfield,
then assign military air traffic
controllers to the evacuation package

Transport military

air traffic controller to ISB

KM/CBR: Possible Future Directions

1. Applications


e
-
Commerce


Decision support systems


Personalized


Knowledge discovery for databases?


Yet KDD stresses need for many
automated
tasks

2. Multimodal systems


e.g., Shimazu: Audio tapes of customer dialogues


Information gathering


Learning assistants

3. Process
-
focused emphases:


Retrieval, adaptation, and composition of processes