intelligent decision support systemsmiquel/idss/idss-part 2-mai-1112.pdf · intelligent decision...
Post on 19-Jun-2020
19 Views
Preview:
TRANSCRIPT
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Intelligent Decision Support Systems
(Part II - EVOLUTION OF DECISION SUPPORT SYSTEMS / INTELLIGENT DECISION SUPPORT SYSTEMS / AN EXAMPLE)
Miquel Sànchez i Marrè, Karina Gibert
miquel@lsi.upc.edu, karina.gibert@upc.edu
http://kemlg.upc.edu/menu1/miquel-sanchez-i-marre
http://kemlg.upc.edu/menu1/karina-gibert
Course 2011/2012
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
PART 2 – EVOLUTION OF DECISION SUPPORT SYSTEMS / INTELLIGENT DECISION SUPPORT SYSTEMS / AN
EXAMPLE
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
EVOLUTION OF DECISION SUPPORT SYSTEMS
Historical Perspective of Management Information Systems
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 20114
Management Information Systems (1) Transaction Processing Systems
early 70s Electronic data processing Large volumes of commercial transactions Batch processing Non interactive
Decision Support Systems (DSS) Started in Middle 70s Interactive systems Supporting business decision making Early DSS
Late 70s Very rudimentary (spredadsheet software)
Advanced DSS 80s Using Operations Research and Management Science Models “what if” analysis
Intelligent Decision Support Systems (IDSS) [90s]
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 20115
Management Information Systems (2)
MANAGEMENT INFORMATION SYSTEMS
TRANSACTION PROCESSING
SYSTEMS
DECISION SUPPORT SYSTEMS
(DSS)
MODEL-DRIVENDSS
DATA-DRIVENDSS
AUTOMATICMODEL-DRIVEN
DSS
TOP-DOWNAPPROACH /
CONFIRMATIVE MODELS
BOTTOM-UPAPPROACH /
EXPLORATIVE MODELS
STATISTICAL / NUMERICAL
METHODS
ARTIF. INTELLIG / MACH. LEARN.
METHODS
INTELLIGENT DECISION SUPPORT SYSTEMS IDSS)
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
EVOLUTION OF DECISION SUPPORT SYSTEMS
Decision Support Systems
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 20117
Need for Decision Making Support Tools
Complexity of the decision making process
Accurate Evaluation of multiple alternatives
Need for forecasting capabilities
Uncertainty
Data Analysis and Exploitation
Need for including experience and expertise (knowledge)
Computational tools: Decision Support Systems (DSS)
Intelligent Computational Tools: Intelligent Decision Support Systems (IDSS)
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 20118
Decision Support SystemsDefinition
A Decision Support System (DSS) is a system under the control of one or more decision makers that assists in the activity of decision making by providing an organised set of tools intended to impart structure to portions of the decision-making situation and to improve the ultimate effectiveness of the decision outcome [Marakas, 1999].
A Decision Support System (DSS) is a system which lets one or more people to make decision/s in a concrete domain, in order to manage it the best way, selecting at each time the best alternative among a set of alternatives, which generally are contradictory
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 20119
Decision Support Systems
DSS in Business and Management Decisions on prices Decisions on products Decisions on marketing Decisions on employees Decisions on strategic planning
DSS in Engineering Decisions on product design Decisions on process supervision / control
DSS in Financial Entities Decisions on loans
DSS in Medicine Decisions about presence/absence of illness Decisions on medical treatments Decisions on specialised medical machine control
DSS in Environmental Systems (EDSS) Decision about reliable and safe environmental system management
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
EVOLUTION OF DECISION SUPPORT SYSTEMS
Advanced Decision Support Systems
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201111
Advanced Decision Support Systems (1)
DSS are a model-based set of procedures for processing data and judgments to assist a manager in his/her decision [Little, 1970]
DSS couple the intellectual resources of individuals with the capabilities of the computer to improve the quality of decisions. It’s a computer-based support for management decision makers who deal with semi-structured problems[Keen & Scott-Morton, 1978]
DSS is a system that is extendable, capable of supporting ad hoc analysis and decision modelling, oriented towards future planning, and of being used at irregular, unplanned intervals [Moore & Chang, 1980]
Definitions
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201112
Advanced Decision Support Systems (2)
DSS enable managers to use data and models related to an entity (object) of interest to solve semi-structured and unstructured problems with which they are faced [Beulens & Van Nunen, 1988]
Main feature of DSS rely in the model component. Formal quantitative models such as statistical, simulation, logic and optimisation models are used to represent the decision model, and their solutions are alternative solutions [Emery, 1987; Bell, 1992]
DSS are systems for extracting, summarising and displaying data [McNurlin & Sprague, 1993]
Definitions
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201113
Advanced Decision Support Systems (3)
Data Analysis
Monitoring / Control
Prediction
Planning
Management
Data Recording / Retrieval
Tasks
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201114
Advanced Decision Support Systems (4)
Statistical /Numerical Models Linear Models, Logistic Regressions Clustering Data Analysis (PCA, DA, SCA, MCA...) Markovian Models
Simulation Models
Control Algorithms
Optimization Techniques Linear Programming Queue Models Inventory Models Transport Models Multiple Criteria Decision Making (MCDM)
Models
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Advanced Decision Support Systems (5)From observations to decisions [adapted from A.D. Witakker, 1993]
15
Observations
Consequences
Recomendations
Predictions
KNOWLEDGE
Understanding
Data
Quantiy of Information
DECISION
INTERPRETATION
High Low
High
Low
Value andRelevance
toDecisions
Observations
Consequences
Recomendations
Predictions
KNOWLEDGE
Understanding
Data
Quantiy of Information
DECISION
INTERPRETATION
High Low
High
Low
Value andRelevance
toDecisions
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Advanced Decision Support Systems (6)Intelligence Density [V. Dhar & R. Stein, 1997]
Intelligence Density : the amount of useful decision support information that a decision maker gets from using the output from some analytic system (IDSS) for a certain amount of time
How much of the IDSS output do you have to examine before you can make a decision of a specified quality ?
How quickly can you get the essence (knowledge) of the underlying data from the IDSS output ?
Conceptually, Intelligence Density can be viewed as the ratio of the number of utiles (utility units) of decision-making power gleaned (quality) to the number of units of analytic time spent by the decision maker.
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201116
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
INTELLIGENT DECISION SUPPORT SYSTEMS
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201118
Intelligent Decision Support Systems (1)
An IEDSS is an intelligent information system that reduces the time in which decisions are made in an environmental domain, and improves the consistency and quality of those decisions [Haagsma & Johanns, 1994]
A DSS is a computer system that assists decision makers in choosing between alternative beliefs or actions by applying knowledge about the decision domain to arrive at recommendations for the various options. It incorporates an explicit decision procedure based on a set of theoretical principles that justify the “rationality” of this procedure [Fox & Das, 2000]
Definitions
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201119
Intelligent Decision Support Systems (2)
The use of Artificial Intelligence tools and models provides direct access to expertise, and their flexibility makes them capable of supporting learning and decision making processes. There integration with numerical and/or statistical models in a single system provides higher accuracy, reliability and utility [Cortés et al., 2000]
Definitions
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201120
Intelligent Decision Support Systems (IDSS) Typology
IDSS in static domains Industrial process design systems
IDSS in dynamic domains Wastewater Treatment Plant Supervisory System
IDSS in dynamic domains with real-time constraints or possible catastrophic consequences Automatic Supervisory System in a medical Intensive
Cure Unit Environmental Emergency management systems
On-line: Emergency coordination systems Off-line: Training systems for emergency management
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201121
Conceptual Components of an IDSS
ARTIFICIAL INTELLIGENCETECHNIQUES
STATISTICAL / NUMERICALMETHODS
GEOGRAPHICAL INFORMATION SYSTEMS /COMPLEMENTARY SYSTEMS
ENVIRONMENTAL / HEALTH /MANAGEMENT / BUSINESS
ONTOLOGIES
TEMPORALREASONING
UNCERTAINTY MANAGEMENT
SPATIALREASONING
ECONOMICALCOSTS
VALIDATION OF IDSS
INTELLIGENT DECISION SUPPORT SYSTEMS
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
IDSS Architecture
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201123
IntelligentDecisionSupportSystems(IDSS) DATA BASE
(TEMPORAL DATA)
SYSTEM / PROCESS
SENSORS
SUBJECTIVEINFORMATION /
ANALYSES
Spatial /Geographical
data
On-linedata
Off-linedata
EXPLANATION ALTERNATIVES EVAL.
PLANNING / PREDICTION / SUPERVISION
REASONING / MODELS’ INTEGRATION
DATA MININGKNOWLEDGE ACQUISITION / LEARNING
AIMODELS
STATISTICALMODELS
NUMERICALMODELS
USER INTERFACE
USER
ON-LINE / OFF-LINE
ACTUATORS
ECONOMICALCOSTS
LAWS ®ULATIONS
Feedback
DAT
A I
NTE
RPRE
TATI
ON
D
ATA
MI
NIN
G
DIA
GN
OSI
S
D
ECIS
ION
SU
PPO
RT
Background/SubjectiveKnowledge
Decision /Strategy
GIS(SPATIAL DATA)
Actuation
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
IDSS Analysis and DesignRequirements, advantages and limitations
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201125
IDSS Requirements
Type of decision problem (policy, operations, resource allocation, etc.)
Domain and scope of the decision problem
Data and knowledge availability
Organizational and structural boundaries
Decision-making process
Impact on and synergy with the existing sytems
Expected consequences of decision execution
Profiles of decision-makers (users of the system)
External constraints and contexts
Objectives of the IDSS
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201126
Kind of Support provided by an IDSS[Marakas, 1999]
Identify the need for decisions
Identify specific problems that require attention
Solve or assist in solving problems
Help compensate for cognitive limitations of human decision makers
Provide assistance in the form of advice, analysis, or evaluation
Enhance user creativity, imagination, or insight
Facilitate interactions in multiparticipant decision maker settings
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
IDSS Requirements and AnalysisThe Stretch Plot: Dimensions for Requirements and Quality Solution[V. Dhar & R. Stein, 1997]
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201127
Accuracy
Explainability
Response Speed
Scalability
Compactness
Flexibility
Easy of Use Embeddability
Tolerance for Complexity
Tolerance for Noisein Data
Tolerance forSparse Data
Learning Curve
DevelopmentSpeed
ComputingEase
Independence from Experts
Model Related
Con
stra
intR
elat
ed
Organization Related
Qua
lity
Rel
ated
QUALITY OF THE SYSTEM ENGINEERING OF THE SYSTEM
QUALITY OF AVAILABLE RESOURCES
LOGISTICAL CONSTRAINTS
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201128
Design and
Building of an IDSS
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
IDSS Validation
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201130
IDSS Validation Criteria (1)Adapted from [Klein & Methlie, 1995]
System Performance Efficiency and response time Data entry Output format Hardware Usage Man-machine interface
Task performance Decision-making time, alternatives, analysis, quality and
participants User perceptions of trust, satisfaction and understanding
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201131
IDSS Validation Criteria (2)Adapted from [Klein & Methlie, 1995]
Business Opportunities Costs of development, operation, and maintenance Benefits associated with increased income and reduced
costs Value to the organization of better service, competitive
advantage, and training
Evolutionary Aspects Degree of flexibility, ability to change Overall functionality of the development tool
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Implementation of an IDSS in a computer. Anexample: Intelligent Environmental System Management
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201133
IntelligentEnvironmental
Decision Support Systems (IEDSS)
DATA BASE(TEMPORAL DATA)
ENVIRONMENTAL SYSTEM / PROCESS
SENSORS
BIOLOGICAL /CHEMICAL /PHYSICALANALYSES
Spatial /Geographical
data
On-linedata
Off-linedata
EXPLANATION ALTERNATIVES EVAL.
PLANNING / PREDICTION / SUPERVISION
REASONING / MODELS’ INTEGRATION
DATA MININGKNOWLEDGE ACQUISITION / LEARNING
AIMODELS
STATISTICALMODELS
NUMERICALMODELS
USER INTERFACE
USER
ON-LINE / OFF-LINE
ACTUATORS
ECONOMICALCOSTS
ENVIRONMENTAL /HEALTH
REGULATIONS
Feedback
DAT
A I
NTE
RPRE
TATI
ON
D
ATA
MI
NIN
G
D
IAG
NO
SIS
D
ECIS
ION
SU
PPO
RT
Background/SubjectiveKnowledge
Decision /Strategy
GIS(SPATIAL DATA)
Actuation
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201134
Intelligent Management of Environmental Systems
Evolutionary Environmental Data Base Management
Intelligent Data Analysis
Acquisition, Learning and Knowledge Management
Reasoning
Cooperation and Integration of several techniques
Decision Making Support
Actuation over the system
Learning
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201135
Operation in a WWTP!
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201136
Proposal for WWTP supervision
HISTORY & ALARMS
LABORATORY
IN-SITU & MICROSCOPICOBSERVATIONS
Operators
MANUALCONTROL
AUTOMATIC CONTROLON DO, RECIRCULATION FLOW,
WASTE FLOW & BY-PASS
Sist. Infom.
ANALITICAL
Laboratory
ON-LINEDATA
Sensors
PLANTMANAGER
ES
SISTEMASUPERVISOR
ACTUATIONPLAN PREDICTION
CBR GPS-X/XT
SUPERVISION
DSS Computer
PLC
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Evolutionary Environmental Data Base Management
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201138
DATA VALIDATION: noise, outliers and missing values
DATA INTEGRATION: homogeneous units and time scale
ON-LINE DATA: 210 digital & 20 analogical signalse.g. pH, flow rates (air, water and sludge), equipment status (valves,pumps, etc.), dissolved oxygen...
OFF-LINE DATA: 158 variablesQUANTITATIVE: e.g. COD, BOD, (V)SS, N, P, Temperature, metals,grease & oils, V30... at different sampling location
QUALITATIVE: e.g. floc characterisation, filamentous bacteria,protozoa and metazoa identification and presence, presence ofbubbles, settling observation...
Data Monitoring
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201139
HSS First level: data gathering
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Intelligent Data Analysis
Acquisition, Learning and Knowledge Management
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201141
Knowledge Acquisition?
Knowledge is the crux matter in natural resources and environmental process management
Decision Making Support must be based on the knowledge
Knowledge acquisition and management from huge amounts of real must be accomplished
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201142
Intelligent Data
Analysis• TRANSITION NETWORKS• TEMPORAL CBR• DYNAMIC CANDID• TIME SERIES
STATISTICAL DATA PROCESSING
AGENT
ENVIRONMENTALDATABASE
FILTERED DATA
CLUSTERINGTECNHIQUES
AGENT
CASE-BASEDREASONING
TECNHIQUESAGENT
REINFORCEMENT /SUPPORT VECTOR
MACHINETECNHIQUES
AGENT
DYNAMICAL ANALYSIS
TECNHIQUESAGENT
EXPERT / USERENVIRONMENTAL
SCIENTIST
GOALSDEFINITION
KNOWLEDGEMANAGEMENT
• Statisticalanalysis
• NN• COBWEB/3• ISODATA• K-MEANS• MARATA
• OPENCASE • CANDID
RECOMMENDER /META-KNOWLEDGE
AGENT
• On-line help• Suggestion of methods
UNIFICATION/INTEGRATION of METHODS/PATTERNS
VALIDATION/INTERPRETATION/DOCUMENTATION
USE ofPATTERNS/KNOWLEDGE
• Prediction• Diagnosis• Planning• Decision support
EXPERT / USERENVIRONMENTAL
SCIENTIST
KNOWLEDGE
RE
CO
MM
EN
DA
TIO
N A
ND
ME
TA
-KN
OW
LK
ED
GE
MA
NA
GE
ME
NT
DA
TA
FILT
ER
ING
KN
OW
LE
DG
E D
ISCO
VE
RY
KN
OW
LE
DG
E M
AN
AG
EM
EN
T
DECISIONTREE
TECNHIQUESAGENT
RULEINDUCTION
TECNHIQUESAGENT
• ID3• CART• C4.5
• CN2• PRISM• RULES• RISE
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201143
STEPS OF THE METHODOLOGY
AbstractionATTRIBUTE
ANDOBSERVATION
SELECTION
CLASSIFICATION
GENERATIONOF
RULESKNOWLEDGE
BASE
Observations:Attribute/Valuesmatrix
Newattributes/observations
Classes Newclassificationparameters
Newattributes/observations
DOMAIN
ATTRIBUTERELEVANCE
MODULE
VALIDATION
E X
P
E
R
TKnowledgePatterns (1): Clustering withLINNEO+
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201144
Knowledge Patterns (2)Q-SP3
SS-AT MICO
COD-SP3 FILAM-PREDOM
C2
<= 929
<= 14
> 0 <= 0
ROTIFER
CIL-CAR T41
<= 1
AMEB COD-AB
<= 1 > 1
COD-SP1 ROTIFER
<= 1
COND-ATC1
> 256 <= 256
C1 C5
<= 898> 898
C1 C19
<= 577> 577> 1
C7 C5
<= 0 > 0
> 1
> 929
> 14
C15 NFILAM
> 0 <= 0
BIODIV-MIC
C8
> 1 <= 1
C19 C1
<= 5 > 5
C5
C1
noc, noc_41,zoo_noc, noc_thio,41, zoo_mico,zoo_41
C16
zoo mico
C9
thio
ESC-B AMEB
<= 1
> 3 <= 3
Q-SP1
COD-EMIS NFILAM
ZOO ASP-AT<= 90 > 90
GPROTO-PREDOM
> 1<= 2
> 1
C8 C1
Sessils,vort_gflag
<=6207
> 1
C14
> 2 <= 2
<= 1
C1121N/THIO
C2 C17
<= 2 > 0
C20
>6207
OPER NOC
C2 AMEB
> 1
C2 C11
<=0 >0
Q-ABC2
> 2
> 1 <= 1
C2 C11
<=11068 >11068
C11<= 1
C4 C3 C2
sessils_rotifer
cap
pflagcrawling,sessils_crawling
T41
> 0 <= 0
OPER
C7
> 1 <= 1
C10
<= 1 >1
C2
(2.0/1.8)
(3.4/2.1) (4.5/1.8)
(66.9/7.2)
(2.0/1.8) (4.2/1.3)
(3.0/1.1)(5.0/2.3)
(3.0/1.1)
(8.0/2.4)(2.0/1.0)
(1.0/1.8) (2.0/1.0)(4.0/2.2)
(5.0/2.3)
C5
cap
(0.0)
(2.0/1.0)
Ameb, sessils_ameb, cilcar,amebteca_rotifers, crawling_cilcar,gflag, crawling_gflag, amebteca,vort_cilcar, vort, cilcar_rotifer,
C4
(0.0) (2.0/1.0) (2.1/1.1) (0.2/0.2) (1.0/0.8) (3.0/2.8)
(2.0/1.8) (4.0/2.2)
(2.0/1.0)
(38.0/2.6)
(7.6/3.5) (2.0/1.0)
(6.0/2.3)
(2.0/1.0) (17.0/1.3)
(3.1/2.1) (9.0/2.4)
(2.0/1.8)
(2.0/1.8)(18.0/2.5)
(18.0/2.5)
Decision Tree Induction
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201145
Knowledge Patterns (3)IF Q-AB < 9083.00AND Q-SP3 < 2227.00AND COND-AT > 847.00AND CIL-CAR < 0.50AND AMEB < 0.50THEN WWTP-problems = c1(47/81)
IF TERB-AB > 123.50AND Q-SP1 > 6223.50AND Q-SP3 > 3467.50AND 33.00 < DQO-AT < 82.00AND COND-AT < 1330.00AND NOC > 0.50AND P-FLAG < 3.50THEN WWTP-problems = c2(38/55)
IF Q-AB < 10924.50AND Q-SP3 > 1030.00AND NOC < 1.50AND MICO > 1.50THEN WWTP-problems = c10(17/17).
IF PH-AB < 7.40AND FILAM-DOMI = thioTHEN WWTP-problems = c9(3/3)
Inference Rule Induction
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201146
Knowledge Patterns (4)...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
..........................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
...................................................................................................................................................................
...............................................................................................................................................................................
...............................................................................................................................................................................
GPROTO-DOMI
FILAM
BIODIV.MIC
COND-AB
Q-AB
NO3-AT
NH4-AT
NFILAM
COD-SP1
Q-SP3
ESC-B
COD-AT
IVF
COD-AB
CM
N
H
?
H
N
Cap
H
L
N
N
?
Cap
Case 100
LH
N
Case 106
N
Case 108
H
N
N
N
Noc
H
L
N
Case 101
N
Case 102
H
N
?
H
?
H
N
N
?
?
Cap
Case 163
N
Case 158
H
Case 155
VH
N
N
Mico
N
N
N
H
Case 157
N
N
Mico
N
?
Cap
N
Case 160H
Case 161
L
L
H
Cap
Case 143
H
Case 153P-flag
Case 144
Amebtecarotifer
N
N
N
N
Thio
L
?
Case 47
N?
Case 46Case 43
Sessils/crawling Crawling
N
H
N
Case 65
H
Noc
Case 178
HN
N
N
N
N
?
N
Case 92Case 179
P-flagVort_gflag
Case 91
Sessils
N
N
Case 172
Zoo/Mico
?
N
N
N
L
Mico
Case 154Case 162
N
H
N
N
H
Cap
Case 107
N
Case 175
?
N
N
Thio
N
L
?
?
Cap
Cap
N
HCase 48
Case 38
H
Case 193
Case 172
Case 159
Case 40
Zoo/41Mico
ZooThio
Case 51
Noc/Thio
LN
......
(L) Low(N) Normal(H) High
Discriminant Tree Induction
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Reasoning
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201148
Expert Knowledge
Prototypical situations
Experts’ domain knowledge
Knowledge acquisition Building-up decision trees Expert interactive knowledge explicitation Machine learning methods
Supervised methods Unsupervised methods
LINNEO+
Inference rules: rule-based reasoning
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201149
STEPS OF THE METHODOLOGY
AbstractionATTRIBUTE
ANDOBSERVATION
SELECTION
CLASSIFICATION
GENERATIONOF
RULESKNOWLEDGE
BASE
Observations:Attribute/Valuesmatrix
Newattributes/observations
Classes Newclassificationparameters
Newattributes/observations
DOMAIN
ATTRIBUTERELEVANCE
MODULE
VALIDATIONE
X
P E
R
TExpert KnowledgeAcquisition: LINNEO+
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201150
SCREEN-KBS-RULE-008 ::IF I-COD 150 mg/L AND I-COD 400 mg/LTHEN conclude that
I-COD-status is normal
BIOLOGICAL-REACTORS-KBS-RULE-023 ::IF DO 0.5 mg/L AND DO 2.5 mg/LTHEN conclude that
DO-level is normal
BIOLOGICAL-REACTORS-KBS-RULE-025 ::IF B-VSS > 2500 mg/LTHEN conclude that
B-VSS-status is high
SECONDARY-SETTLER-KBS-RULE-031 ::IF O-COD 75 mg/LTHEN conclude that
O-COD-status is high
SUPERVISORY-KBS-RULE-002 ::IF I-COD-status is normal AND O-COD-status is highTHEN conclude that
COD Removal efficiency is low
SUPERVISORY-KBS-RULE-037 ::IF COD Removal efficiency is low AND B-VSS-status is highTHEN conclude that
SIDESTREAMS situation is VERY-POSSIBLE
LOCAL-DIAGNOSIS RULES
COMBINATION RULES
Diagnosis Rules
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201151
SHELLMecanismed’inferència
USUARI
Interfíciegràfics/menus
BASE DE CONEIXEMENTReglesÀrbres de decisió
DADES EXTERNESSensorsLaboratoriObservacions
Aplicacionsexternes
SBC
Knowledge-Based System
Architecture of a Knowledge-Based System
Knowledge Acquisition
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201152
Experiential Knowledge
Ideosyncratic situations
Practical experiented knowledge
Cases or experiences: case-based reasoning
Learning
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201153
Case-Based Reasoning
CASELIBRARY
INDEX
NEWCASE
SUCCESSFUL/FAILURE
SOLUTION
BEST CASE
ADAPTEDSOLUTIONDOMAIN
KNOWLEDGE
RETRIEVAL
RETRIEVEDCASES
SIMILARITY ASSESMENT
EVALUATION
ADAPTATION
LEARNING
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201154
Case Structure
( :identifier CASE-31:situation-description ( (Water-inflow 22,564 m3/day)
(Inflow-Biological-Oxygen-Demand 192.9 mg/L). . . )
:diagnostics STORM-SITUATION:actuation-plan( (1 Close Purge-flow)
(2 Maintain-DO-predictive-control). . . )
:case-derivation CASE-11:solution-result SUCCESS:utility-measure 0.63:distance-to-case 0.1305 )
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201155
Atribut Ordre Pes RangBaix (B) Normal (N) Alt (A) Molt Alt (MA)
DQO-Aigua Tractada (DQO-AT)Index Volumètric Fangs (IVF)Presència Escumes Aireació (ESC-B)DQO-entrada (DQO-AB)Biodiversitat microfauna (BIODIV-MIC)Conductivitat-entrada (COND-AB)Cabal-entrada (Q-AB)DQO-Sortida Primari#1 (DQO-SP1)Cabal-Sortida Primari#3 (Q-SP3)Nitrat-Aigua Tractada (NO3-AT)Amoni-Aigua Tractada (NH4-AT)Número Filamentosos (NFILAM)Càrrega màsssica (CM)SST-Aigua Tractada (SST-AT)Aspecte-Aigua Tractada (ASP-AT)
1234
6789
1011121415--
8.0109.05.0
8.05.06.05.01.077888
10
( 0 - 40 )( 0 - 275 )
-( 0 - 400 )
( 0 – 2.1 )( 0 - 900 )( 0 - 4500 )( 0 - 250 )( 0 - 1000 )
( 0 - 1 ) ( 0 - 8.0 )
-( 0 - 0.25 )( 0 – 10)( 0 – 1 )
( 40 - 70 )( 275 – 500 )
( 0 – 2.5 )( 400 – 950 )
( 2.1 – 4.5 )( 900 – 1800 )
( 4500 – 17000 )( 250 – 550 )
( 1000 – 5500 )( 1 - 6 )
( 8.0 - 25 )( 0 – 4.25)
( 0.25 - 0.35 )( 10 – 20 )( 1 – 2 )
( 70 – 150 )( 500 – 1100 )
( 2.5 – 5)( 950 – 1500 )
( 4.5 – 8 )( 1800 – 2500 )(17000–30000)( 550 –1500 )
( 5500 – 11000 )( 6 – 20 )( 25 - 50 )
( 4.25 – 8 )( 0.35 - 1.5 )( 20 – 50 )( 2 – 5 )
( 150 – 500 )--
( 1500 – 3000 )
-(2500-10000)
(30000-50000)-
(11000-30000)( 20 – 60 )( 50 –100 )
-( 1.5 – 5.0 )( 50 – 200 )
-Espècie o Grup protozoo dominant(GPROTO-DOMI) 13 8 Sessils Crawling P-flag Rotífers Cil-Car Amebes Cap
Filamentós Predominant (FILAM) 5 10 Zooglea Microthrix 021/thiotrix 0041 Noc/0041 Nocardia Cap
Case-Based Reasoning System
CASE LIBRARY SEEDEDWITH 24 CASES
# C A S D IA S IT U A C IÓ # C A S D IA S IT U A C IÓ
1 2 7 /2 /9 6 N o rm a l H iv e rn 1 3 8 /8 /9 6 N o rm a l E s tiu
2 3 /1 /9 7 P lu ja 1 4 1 0 /1 /9 7 P lu ja
3 9 /1 2 /9 6 T em p es ta 1 5 5 /9 /9 6 D es f lo cu la c ió
4 3 /7 /9 7 N o rm a l E s t iu 1 6 2 8 /1 1 /9 6 N o rm a l H iv ern
5 1 2 /9 /9 6 X o c d e C à r reg a 1 7 2 5 /8 /9 6 N o rm a l E s tiu
6 1 1 /1 2 /9 6 B a ix a C à r reg a 1 8 2 7 /1 1 /9 7 M ic o th r ix iZ o o g lea
7 6 /1 1 /9 7 D es n itr if ica c ió 1 9 2 8 /1 0 /9 7 X o c d e C lo ru rs
8 3 /1 0 /9 6 D eca n .1 ª d e fic ien t 2 0 3 /6 /9 7 F o a m in g p e rN o ca rd ia
9 7 /5 /9 6 B u lq u in g p e rT h io th r ix 2 1 2 3 /1 /9 7 T erb o les a p er p -
fla g e lta s
1 0 2 5 /3 /9 7 F o a m in g p e rM ic o th r ix 2 2 2 0 /1 1 /9 7 N itr ifica c ió
1 1 2 3 /1 /9 6 F o a m in g p e rM ic o th r ix 2 3 1 6 /1 2 /9 7 T em p es ta
1 2 1 5 /4 /9 7 F o a m in g 2 4 2 5 /4 /9 6T ra n s ic ió d es itu a c ió n o rm a l aB u lq u in g
B
B
N
A NA A
N
N
Cas 14
N
A A N
N
Noc/Thio
Cas 11
Cas 21Cap
B
Cas 2
Cas 18
N
NBN
N B
Cas 3B
N
N
N
Cas 22
B
N
Cas 6 Cas 16B N Cas 5
A
ZooCas 19
N
Cas 12 Cas 10Cap
N
Cas 25
N
Crawling
Mico Noc
Cas 13 Cas 7N A B
A
Noc/Thio
?
Cas 1
Cas 15N
B
Cas 8 A
Mico
A
N
N
NCas 9
?
?
Cil-Car
A
CASE LIBRARY SEEDED WITH 24 CASES
Cas 23
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201156
Ontologies
Issues To obtain a unified (but multilingual), complete and
coherent terminology for a domain Knowledge reuse and sharing To help in the management of problematic situations:
impasses
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201157
Ontologies
Concepts organized in taxonomies, linked by relations and conforming to axioms.
Axioms can be very general, e.g.: “foxes eat fish”
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Integration and cooperation of differenttechniques
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201159
DAI-DEPUR: a hybrid AI framework (1)
DOMAIN THEORY / MODELS PROCESSES / METHODS
IDENTIFIED SITUATION(S)
OFF-LINE VALUESDATA GATHERING
ADOPTED SOLUTION
OPERATOR’S VALIDATION
COMBINATION
MODEL-BASED REASONING
RULE-BASED REASONING
CASE-BASED REASONING
KNOWLEDGE ACQUISITION
LEARNING
EXPERT KNOWLEDGE
EXPERIENTIAL KNOWLEDGE
NUM. CONTROL KNOWLEDGE
PREDICTIVE KNOWLEDGE
SPECIFIC SITUATION(S)
GENERAL SITUATION(S)
PROPOSED SOLUTION(S)
EXPERIENTIAL/ EXPERT
ACTUATION
NUMERICALCONTROL
ACTUATION
ACTUATION
VALIDATION
SUPERVISION
PREDICTION
DIAGNOSIS
ADAPTATION
EVALUATIONDATAON-LINE SENSORS
KNOWLEDGE / EXPERTISE
SITUATIONS
PLANS
SUPERVISORY CYCLE
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201160
Numerical control knowledge
Normal situations
Numerical computation
Mathematical control algorithm For instance, DO (Dissolved Oxygen) Predictive control
algorithm for WWTP
Optimisation of process
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201161
Operation of the Supervisory Cycle
Data Base Information Gathering
The Supervisory Module is activated at the beginning or the end of the day or when
the user requirement
File Importation with average values of the day before
ACTIVATION of both systems in PARAL·LEL
CASE-BASED SYSTEM RULE-BASED SYSTEM
WWTP MODELS: GPS-X / T.N.
PREDICTION
ADAPTACIÓ
RECUPERACIÓ
Bibliotecade casos
Noucas
Millorcas
Casadaptat
Diagnosis & actuation proposal from adapted CASE
Information Combination and situation inferenceSUPERVISORY MODULE
Diagnosis & Actuation inferred
REGLES i ARBRESde DECISIÓ
ACTUATIONPLAN proposal
WWTP
APRENENTATGE
AVALUACIÓ
PLANT MANAGER
EXPERT, EXPERIENTIAL ACTUATION OR by NUMERICAL CONTROL
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Decision Making Support
Transition Networks, Numerical simulations
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201163
Predictive knowledge
Transitions between situations
Practical experiented knowledge
Transition networks: model-based reasoning
Prediction
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201164
The Transition Network Model
S1
S3
S2
S0
I4, C1
I2, C4
I1, C5
I2, C3
I3, C6
I4, C1
I2, C5
I2, C1
I2, C4
I3, C6
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201165
Part of a transition networkBad sedimentability transitions
S1S3
(I1,C1)
(I1,C1)
(I1,C1)
(I1,C1)S2
(I1,C2)
(I1,C1)
(I2,C1)
(I2,C1)
(I1,C2)
(I4,C1)(I2,C1)
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Actuation over the system
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201167
Actuation
Off-line Actuation Operator Validation
On-line Actuation Automatic Actuation
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Learning
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201169
Learning by Experience
Learning new problems and new solutions
Improve the problem solving process
Improve the diagnosis process
Avoid making the same mistakes
System increases performance with time
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
PRINCIPLES OF ARTIFICIAL INTELLIGENCE
Fundamentals of AI
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
71
Artificial Intelligence (1)
E. Charniak & D. McDermott, 1985 Artificial Intelligence is the study of mental faculties through
the use of computational models.
E. Rich & K. Knight, 1991 Artificial Intelligence (AI) is the study of how to make
computers do things, which, at the moment, people do better
L. Steels, 1993 Artificial Intelligence is a scientific research field concerned
with intelligent behaviour.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
72
Artificial Intelligence (2) OMNI:
What is this the main goal of AI?
Herbert A. Simon (1994):AI can have two purposes. One is to use the power of computers to augment human thinking, just as we use motors to augment human or horse power. Robotics and expert systems are major branches of that. The other is to use a computer's artificial intelligence to understand how humans think. In a humanoid way. If you test your programs not merely by what they can accomplish, but how they accomplish it, they you're really doing cognitive science; you're using AI to understand the human mind.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
73
Artificial Intelligence (3)
Artificial Intelligence is the study of the possible or existing mechanisms –in human or other beings–providing such behaviour in them that can be considered as intelligence, and the simulation of these mechanisms, named as cognitive tasks, in a computer through the computer's programming.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
74
Artificial Intelligence Areas by application field Puzzle Resolution
Automatic Theorem Proving/Logic Programming
Game Theory
Medical Diagnosis
Machine Translation
Symbolic Mathematics and Algebra
Robotics
Fault Diagnosis
Text Understanding and Generation
Monitoring and Control Systems
e-Commerce
Business Intelligence
Intelligent Decision Support Systems
Recommender Systems
Intelligent Web Services
Social Networks Analysis
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
75
Artificial Intelligence Areas by Cognitive Tasks
Vision and Perception
Natural language understanding
Knowledge acquisition
Knowledge representation
Reasoning
Search
Planning
Explanation
Learning
Motion
Speech and Natural language generation
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
76
AI Goals The construction of Intelligent Systems The construction of such intelligent systems or intelligent
agents means that the agents must show an intelligent behaviour like: Autonomy Learning skills Communication abilities Coordination abilities Collaboration abilities with other agents, either humans or
artificial ones.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
77
Human Beings or “Intelligent” Agents BehaviourEXTERNAL STIMULIVisionNatural Language Understanding
INTERNAL STIMULI
INTERNAL COGNITIVE TASKS INTERNAL REACTION ACTIONSReasoning, Know. Acquisition, Search, Learning, …
Human being Goals Knowledge
EXTERNAL REACTION ACTIONSMotionNatural Language Generation
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
78
Intelligent Agents
An intelligent agent is a computer system that is capable of flexible autonomous action in order to meet its design objectives (Wooldridge and Jennings, 1995).
Agents must be: Responsive: agents should perceive their environment and
respond in a timely fashion to changes that occur in it, Proactive: agents should be able to exhibit opportunistic,
goal-directed behaviour and take the initiative where appropriate
Social: agents should be able to interact, when they deem appropriate, with other artificial agents and humans in order to complete their own problem solving capabilities and to help others with their activities.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
79
ExamplesAdapted from [Russell & Norvig 2000]
Agent Type Goals / Performance Measure Environment Sensors Actuators
Medical Diagnosis System Healthy patient, Reduced costs
Patient, hospital, staff Keyboard entry of symptoms, findings, patient’s answer
Display of questions, tests, diagnoses, treatments, referrals
Satellite image analysis system
Correct image categorization
Images from orbiting satellite
Colour pixel arrays Display of scene categorization
Internet Softbot Summarise relevant information
Users, the web Web pages, text, links Interesting topic detection in web pages, going to other web pages
Machine Translation system
Text translation from one language to another
Text, Linguistic knowledge Text Sentence translation, Semantic interpretation of paragraphs
MARS Pathfinder Maximum information gathering from Mars
Mars surface, Earth signals Telemetry, Mars environment images, Velocity and balance information
Forward, backward, accelerate, braking, turning, take one photo
Refinery controller Purity, yield, safety Refinery, operators Temperature, pressure, chemical sensors
Valves, pumps, heaters, displays
Interactive English tutor Student’s score on test satisfactory
Set of students, testing agency
Keyboard entry Display of exercises, suggestions, corrections
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
80
Kinds of Agents
Cognitive or deliberative agents: They are based in a deep knowledge representation. Rational actions are implemented through inference
processes by means of rule sets. For instance, a medical diagnosis system built up on a
deep medical knowledge, which provides a deliberative reasoning process.
Reactive or behavioural agents: They are a set of behaviours, which interact and evolve
to reach their goals. They are prepared to react to external stimuli. For instance, a mobile robot, which interacts with the
environment showing emergent behaviours.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
81
A Cognitive Agent: MYCIN
Medical Diagnosis Infeccious Breathing diseases Great Medical Knowledge Compilation process
Actuation Questions and answers with the doctor
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
82
MYCIN: Example Dialogue (I)
>What is the patient’s name? John Smith >Age? He is 55 {Natural Langiage Interpretation}>Have you obtained positive cultures indicating general>type? {Evidence Caption}Yes>Let’s call the most recent culture CULTURE-1. Fromwhat site was CULTURE-1 taken? From the blood>When? June, 21, 2001>Let’s call the first significant organism from this blood>culture ORGANISM-1. Do you know the identity of>ORGANISM-1? No
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
83
MYCIN: Example Dialogue (II)
>Is ORGANISM-1 a rod or a coccus or something else? Rod {Cause Discrimination}>What is the gramstain of ORGANISM-1? Gramnegative> Has John Smith a previous history of alcoholism?No> Is there evidence that the infection has hospitalary origin?Yes> My therapy recommendations will be based on the>following possible identities of the organism(s) that seem>to be significant:the identity of ORGANISM-1 may be
PSEUDOMONAS
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
84
MYCIN: Example Dialogue (III)
>the identity of ORGANISM-2 may be KLEBSIELLA
the identity of ORGANISM-3 may be
ENTEROBACTER
>My preferred therapy recommendation is as follows:
>Give the following in combination:
>GENTAMYCIN
>Dose: 1.7 mg/kg Q8H - IV or IM
>Comments: Modify dose in renal >failure
>CARBENICILLIN {etc.}
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
85
A Reactive Agent: Marvin
Data Monitoring Repeated Supervision of many parameters Alarm Categorization
Actuation / Reaction Data Anomalies Detection Sending Messages to Operators Alarm Detection
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
86
AI differences with Conventional Programming
ARTIFICIALINTELLIGENCE
CONVENTIONALPROGRAMMING
Symbolic processing Numerical processing
Complex Domains “Normal” Domains
Heuristic search/resolution Algorithmic search/resolution
Approximate/suboptimal solutions “Exact”/Optimal solutions
Approximate/Uncertain Information Exact/certain Information
Functional/Logic Programming Procedural/Object-Oriented Programming
https://kemlg.upc.edu
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
PRINCIPLES OF ARTIFICIAL INTELLIGENCE
AI Paradigms
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
88
AI Paradigms or Approaches
The deliberative/symbolic paradigms Concerned on the processing of symbols rather than
numerical values. Use a latent reasoning mechanism. Most of them are Cognitive-inspired approaches.
The reactive/subsymbolic paradigms Concerned about more numerical computations and providing
nice and intelligent optimizations schemes or function approximation schemes.
No evident reasoning mechanisms are used. Most of them are Bio-inspired approaches.
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Deliberative Approaches (1)
Logic paradigm: based on representing the knowledge about the problem and the domain theory through logical formulas. The main reasoning mechanism is the automatic theorem
proving using the automated resolution process set by Robinson. It is a very general mechanism.
Major techniques are based on Logic Programming.
Heuristic search and planning paradigm: it is based in searching within a space of possible states, starting from the initial state to a final state, where the problem has been solved. The state space is a graph structure to be intelligently explored. Most commonly used techniques are the A* algorithm and
other heuristic search approaches.
© Miquel Sànchez i Marrè, KEMLG, 201189
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Deliberative Approaches (2)
Knowledge-Based paradigm: this kind of approach tries to get benefit from the particular knowledge of a concrete domain, which is normally used by experts when facing the problems to be solved. This knowledge is encoded in what has been named as a
Knowledge Base. Most common knowledge bases are implemented as inference rules (IF <conditions> THEN <actions>). The knowledge is explicitly represented by the inference rules.
Main examples of this paradigm are the Expert Systems and the Intelligent Tutoring Systems.
The reasoning mechanisms are the forward and backward reasoning engines.
The technique used here is commonly known as Rule-Based Reasoning (RBR), or sometimes also known as Knowledge-Based System (KBS).
© Miquel Sànchez i Marrè, KEMLG, 201190
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Deliberative Approaches (3)
Model-Based paradigm: this approach is very similar to the Knowledge-based, because the knowledge of a particular domain is used. The difference relies on the fact that the knowledge is implicitly encoded in some kind of model. Most common approaches are causal models reflecting the
causal relationships among several components of a system, or qualitative models reflecting the qualitative relationships among several attributes which are characterizing the domain.
The reasoning process is done through some kind of interpreter of the model and its component relationships.
Model-Based Reasoning (MBR) and Qualitative Reasoning are major techniques using this approach
© Miquel Sànchez i Marrè, KEMLG, 201191
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Deliberative Approaches (4)
Experience–Based paradigm: this approach tries to solve new problems in a domain using the solution given in the past to a similar problem in the same domain (analogical reasoning). Thus, the solved problems constitute the “knowledge” about the domain. As more experienced is the system better performance achieves,
because the experiences (cases or solved problems) are stored in the Case Base.
This way the system is continuously learning to solve new problems.
The technique used in this approach is known as Case-Based Reasoning (CBR) or Instance-Based Reasoning.
© Miquel Sànchez i Marrè, KEMLG, 201192
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Reactive Approaches (1)
Connectionism paradigm: this approach is inspired by the biological neural networks which are in the brain of many living beings. The model of an Artificial Neural Network (ANN) mimics the
biological neural networks with the interconnection of artificial neurons.
The ANN will produce an output result, as an answer to several input information from the input layer neurons, emulating the neural networks of the brain which propagate signals among all the neurons interconnected.
ANNs are general approximation functions very useful in nonlinear conditions
© Miquel Sànchez i Marrè, KEMLG, 201193
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Reactive Approaches (2)
Evolutionary Computation paradigm: Evolutionary computation is a bio-inspired approach mimicking selection natural process in biological populations. It uses iterative progress, such as growth or development in a
population. This population is then selected in a guided random search being able to use parallel processing to achieve the desired end.
Such processes are often inspired by biological mechanisms of evolution.
Evolutionary computation provides a biological combinatorial optimization approach.
© Miquel Sànchez i Marrè, KEMLG, 201194
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Reactive Approaches (3)
Uncertainty reasoning paradigms: A Bayesian network, belief network or directed acyclic
graphical model Is a probabilistic graphical model that represents a set of
random variables and their conditional dependencies via a directed acyclic graph (DAG).
They provide an inference reasoning mechanism to obtain the new probability values of any variable within the network after some new evidences are known
Fuzzy logic systems Systems based on fuzzy logic and possibilistic theory to
model the vagueness and imprecision concepts. The mathematical possibilistic model assigns a possibility
value to each element which is evaluated regarding whether it belongs to a set. Values are evaluated in terms of logical variables that take on continuous values between 0 and 1.
© Miquel Sànchez i Marrè, KEMLG, 201195
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
Reactive Approaches (4)
Other optimization paradigms: there are several optimization techniques approaches ranging from Collective problem solving techniques named as
“Swarm Intelligence” such as Ant colony optimization Swarm particle optimization Etc.
Until another simpler optimization techniques using several meta-heuristics
© Miquel Sànchez i Marrè, KEMLG, 201196
Inte
llige
nt D
ecis
ion
Supp
ort S
yste
ms
© Miquel Sànchez i Marrè & Karina Gibert, KEMLG, 201197
Miquel Sànchez i Marrè, Karina Gibert(miquel@lsi.upc.edu, karina.gibert@upc.edu)
http://kemlg.upc.edu/
top related