Академический Документы
Профессиональный Документы
Культура Документы
Contents
Introduction: OntoServ.Net Global HealthCare Environment for Industrial Devices;
Bayesian Metanetworks for Context-Sensitive
Industrial Diagnostics;
Temporal Industrial Diagnostics with
Uncertainty;
Dynamic Integration of Classification
Algorithms for Industrial Diagnostics;
Industrial Diagnostics with Real-Time NeuroFuzzy Systems;
Systems
Conclusion.
Vagan
Terziyan
Andriy
Zharko
Oleksandr
Kononenko
Oleksiy
Khriyenko
Host Agent
Maintenance
Service
Service Agents
OntoServ.Net Challenges
New group of Web service users smart industrial
devices.
devices
Internal (embedded) and external (Web-based) agent
enabled service platforms.
platforms
Mobile Service Component
Component concept supposes that any
service component can move, be executed and learn at
any platform from the Service Network, including service
requestor side.
Semantic Peer-to-Peer concept for service network
management assumes ontology-based decentralized
service network management.
3. Wait a bit, I
will give you
some pills
2. I think you
should stop drink
beer for a while
GUN Concept
2. I have some
pills for you
1. I feel bad,
temperature 40, pain in
stomach, Who can
advise what to do ?
= GUN Resource
GUN
OntoShells
OntoAdapters
A. Zharko
Vagan Terziyan
Oleksandra Vitko
P(Y|X)
Y
P(Y)-?
Conditional (in)dependence rule
P ( X 1 , X 2 ,..., X n ) P ( X i | Parents ( X i ))
i 1
P(Y y j , X xi ) P( X xi ) P(Y y j | X xi )
P(Y y j ) P ( X xi ) P(Y y j | X xi )
P( X xi | Y y j )
P( X xi ) P(Y y j | X xi )
P(Y y j )
Bayesian rule
Machine
emission
Environment
Sensors
x1
x2
x3
predictive attributes
x4
x5
x6
x7
contextual attributes
x1
x2
x3
x4
xk
x6
x7
contextual attributes
predictive attributes
Assume conditional
dependence between
predictive attributes
(causal relation between
physical quantities)
x5
xt
xr
some contextual
attribute may effect
directly the conditional
dependence between
predictive attributes but
not the attributes itself
P(P(Y|X))
P(Z)
P(Y|X)
pk(Y|X)
P(P(Y|X)|Z)
P(Y)-?
P (Y y j ) { pk (Y y j | X xi ) P( X xi )
k 1 i 1
[ P ( Z z m ) P( P(Y | X ) pk (Y | X ) | Z z m )]}
m1
x2
x3
x4
xk
x7
xt
P(X)
x1 x2 x3 x4
x6
contextual attributes
predictive attributes
Assume some predictive
attribute is a random
variable with appropriate
probability distribution
for its values
x5
some contextual
attribute may effect
directly the probability
distribution of the
predictive attribute
Probability
P(P(X))
P(X)
pk(X)
P(Z)
P(Y|X)
P(P(X)|Z)
P(Y)-?
Y
P(Y y j ) {P (Y y j | X xi ) pk ( X xi )
k 1 i 1
[ P( Z z m ) P( P( X ) pk ( X ) | Z z m )]}
m 1
Contextual level
A
Q
B
Q
B
Predictive level
Contextual level
A
Q
B
Predictive level
A
Q
B
Terziyan V., Vitko O., Probabilistic Metanetworks for Intelligent Data Analysis, Artificial Intelligence,
Donetsk Institute of Artificial Intelligence, Vol. 3, 2002, pp. 188-197.
Terziyan V., Vitko O., Bayesian Metanetwork for Modelling User Preferences in Mobile Environment, In:
German Conference on Artificial Intelligence (KI-2003), Hamburg, Germany, September 15-18, 2003.
A
B
Predictive level
xm
P(P(Xn| Xm))
P(Xn| Xm)
P1(Xn|Xm) P2(Xn|Xm) P3(Xn|Xm)
P(P(Xr| Xk))
P(Xr| Xk)
P1(Xr|Xk) P2(Xr|Xk)
xk
xr
P(X)
A
P(P(B|A))
P(P(Y|X))
P(B|A)
P(Y|X)
pk(Y|X)
pr(B|A)
B
P(Y)-?
P(P(Y|X)|P(B|A))
P(Y y j ) { pk (Y y j | X xi ) P( X xi )
i
[ P( P(Y | X ) pk (Y | X ) | P( P( B | A) pr (Y | X )) P( P( B | A) pr ( B | A))]}.
r
Contextual level
P(A)
P(X)
Predictive level
P(P(A))
b)
pr(A)
P(A)
P(P(X)|P(A))
P(P(X))
P(X)
pk(X)
Contextual level
P(A)
A
Y
P(P(A))
A
P(Y|X)
Predictive level
d)
P(X)
X
P(P(Y|X))
P(A)
P(Y|X)
pk(Y|X)
pr(A)
P(P(Y|X)|P(A))
P(Y)-?
P(A)
P(B)
Contextual level
P(P(A))
A
P(X)
X
P(A)
P(P(Y|X))
P(Y|X)
P(Y|X)
Predictive level
A
B
f)
pr(A)
P(P(Y|X)|P(A))
B
pk(Y|X)
Y
P(B)
P(P(A)|P(B))
ps(B)
P(P(B))
P(Y)-?
Predictive level
Probability
(X)
P(X)
P(Y|X)
P(Y)-?
Y
P(Y )
Probability to have this model
is:
P((X)=no)= 1-X
We consider relevance as
a probability of importance
of the variable to the
inference of target attribute
in the given context. In
such definition relevance
inherits all properties of a
probability.
1
P(Y | X ) [nx X P( X ) (1 X )].
nx X
P(X)
P0(Y)
X
Probability to have this
model is:
P((X)=yes)= X
P(Y|X)
Y
Y
P(A)
(X)
P(X)
P((X)|(A))
P(Y|X)
Y
P (Y )
In a relevance network
the relevancies are
considered as random
variables between
which the conditional
dependencies can be
learned.
P(Y)-?
1
{P (Y | X ) [nx P ( X ) P( X | A ) P( A ) (1 X )]}.
nx X
A
(X)
Probability
Relevance
P(X)
(Z)
Probability
P(Z)
Probability
Probability
P(X)
P(Z)
P(Y|X,Z)
Y
P(X)
X
P(Y|X)
P(Y)-?
Probability
Probability
P(Z)
P((X)=no)P((Z)=no) =
= (1-X)(1-Z)
P((X)=no)P((Z)=yes) =
= (1-X)Z
P(Y|Z)
P((X)=yes)P((Z)=no) =
= X(1-Z)
P(Y|X,Z)
Probability
P0(Y)
Y
(X)
Probability
Relevance
P(X)
(Z)
Probability
P(Z)
P(Y|X,Z)
Y
P(Y)-?
nx
nz
P(Y ) X Z P (Y | X xi , Z z k ) P( X xi ) P( Z z k )
i 1 k i
1 nx nz
X (1 Z ) P(Y | X xi , Z z k ) P ( X xi )
nz i 1 k i
1 nx nz
(1 X ) Z P(Y | X xi , Z z k ) P( Z z k )
nx i 1 k i
nx nz
1
(1 X ) (1 Z )
P (Y | X xi , Z z k ),
nx nz i 1 k i
(X2)
Probability
Probability
Probability
P(XN)
P(X2)
Relevance
P(X1)
X2
(XN)
X1
XN
Relevance
(X1)
P(Y|X1,X2,,XN)
Y
P(Y)-?
XN = P((XN) = yes);
Goal: to estimate P(Y).
(X2)
Probability
Probability
Probability
P(XN)
P(X2)
Relevance
P(X1)
X2
(XN)
X1
XN
Relevance
(X1)
P(Y|X1,X2,,XN)
Y
P (Y )
1
N
nxs
s 1
XN
P(Y)-?
nxr
Xr
P ( Xr )
(1
q ( ( Xq )"no ")
Xq
)]
Q
B
Predictive level
b)
c)
Q
B
Relevance
level
Combined Bayesian
Metanetwork
Contextual level A
Contextual level B
Predictive level
Learning Bayesian
Metanetworks from Data
Learning Bayesian Metanetwork structure
(conditional, contextual and relevance
(in)dependencies at each level);
Learning Bayesian Metanetwork parameters
(conditional and unconditional probabilities
and relevancies at each level).
Vitko O., Multilevel Probabilistic Networks for Modelling Complex
Information Systems under Uncertainty, Ph.D. Thesis, Kharkov National
University of Radioelectronics, June 2003. Supervisor: Terziyan V.
2.
Vagan Terziyan
Vladimir Ryabov
N2
N1
N3
DN ,S n
DN ,S1
DN ,S 2
S
S1
N5
N4
S2
Sn
Temporal scenarios
Estimation
Recognition
Diagnosis
Relational
network
Industrial object
DB of
scenarios
Learning
Event 2
Ryabov V., Handling Imperfect Temporal Relations, Ph.D. Thesis, University of
Jyvaskyla, December 2002. Supervisors: Puuronen S., Terziyan V.
Event 2
<
>
R(Event 1,Event 2)
rb,a = ~
ra,b
ra,b
rb,c
ra,b
r2a,b
r1a,b
Inversion
a
ra ,b r1a ,b r 2 a ,b
Sum
A before (b) B
B after (bi) A
A meets (m) B
B met-by (mi) A
A overlaps (o) B
B overlapped-by (oi) A
A starts (s) B
B started-by (si) A
A
B
A during (d) B
B contains (di) A
A
B
A finishes (f) B
B finished-by (fi) A
A
B
A equals (eq) B
B equals A
interval 2
a1;
a2;
a13;
Sensor 1
Estimation of
temporal
relations
between
symptoms
Sensor 2
Industrial object
Sensor 3
Relational network
representing the
particular case
Object A
N1
Object C
N2
Scenario S
1. for i=1 to n do
2. for j=i+1 to n do
3.
4.
begin
5.
for g=1 to n do
6.
7.
8.
9.
end
10.
else go to line 2
N3
Generating the
temporal
scenario
for Failure X
DB of
scenarios
Estimation
Recognition
Diagnosis
Relational
network
Industrial object
DB of
scenarios
Learning
D N,S
w d
i 1
m
w
i 1
Probability
value
o
m
fi
di
si
eq
d
s
oi
m
i
bi
1
i 1
i
e
Bal(RA,B) =
A, B
12 i 0
wb =0
Balance point for
RA,B
weq
=0.5
wf
=0.75
Balance point for
RC,D
wbi
=1
2.
3.
Vagan
Terziyan
Terziyan V., Dynamic Integration of Virtual Predictors, In: L.I. Kuncheva, F. Steimann,
C. Haefke, M. Aladjem, V. Novak (Eds), Proceedings of the International ICSC Congress
on Computational Intelligence: Methods and Applications - CIMA'2001, Bangor, Wales,
UK, June 19 - 22, 2001, ICSC Academic Press, Canada/The Netherlands, pp. 463-469.
The Problem
During the past several years, in a variety of
application domains, researchers in machine
learning, computational learning theory, pattern
recognition and statistics have tried to combine
efforts to learn how
ensemble of classifiers.
The primary goal of combining several classifiers is to
obtain a more accurate prediction than can be
obtained from any single classifier alone.
Combination
Selection
Decontextualization
Global
(Static)
Local
(Dynamic)
Local
Global
(Virtual
(Voting-Type) Classifier)
Learning Environment
Predictors/Classifiers
P1
P2
...
yt
Pn
Virtual Classifier
Virtual Classifier is a group of seven cooperative agents:
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP,
TI
,
FS,
DE,
CL
Team Instructors
Classification Team
TC - Team Collector
FS - Feature Selector
TM - Training Manager
DE - Distance Evaluator
TP - Team Predictor
CL - Classification Processor
TI - Team Integrator
Classification Team:
Feature Selector
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP,
TI
,
FS
,
DE,
CL
Team Instructors
Classification Team
FS - Feature Selector
Feature Selector:
Sample Instances
r yr
Feature Selector
'r yr , 'r r
Classification Team:
Distance Evaluator
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP,
TI
,
FS,
DE
,
CL
Team Instructors
DE - Distance Evaluator
Classification Team
i d ( xi , yi )
D( X , Y )
i , xiX , yi Y
where:
0, if xi yi
if i th attribute is nominal -
1, otherwise
d ( xi , yi )
else : xi yi
rangei
d (red, yellow) = 1
Distance Evaluator:
measures distance between instances based on
x j1, x j 2 ,..., x jm
xi1 , xi 2 ,..., xim
their numerical or nominal attribute values
Distance Evaluator
dij
Classification Team:
Classification Processor
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP,
TI
,
FS,
DE,
CL
Team Instructors
Classification Team
CL - Classification Processor
Classification Processor:
predicts class for a new instance based on its selected
features and its location
x relatively
, x ,..., x to sample instances
i1
Sample Instances
i2
im
Feature
Selector
Classification
Processor
Distance
Evaluator
yi
Team Instructors:
Team Collector
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP,
TI
,
FS,
DE,
CL
Team Instructors
Classification Team
Team Collector
completes classification teams for future training
Distance Evaluation
functions
Classification
rules
Feature Selection
methods
Team Collector
FSi
DEj
CLk
Team Instructors:
Training Manager
Constant
Team
Members
Elective
Team
Members
TC,
TM
,
TP,
TI
,
FS,
DE,
CL
Team Instructors
Classification Team
Training Manager
trains all completed teams on sample instances
Training Manager
Sample Instances
xr1 , xr 2 ,..., xrm yr
FSi1
DEj1
CLk1
FSi2
DEj2
CLk2
FSin
DEjn
CLkn
Classification Teams
Sample Metadata
Team Instructors:
Team Predictor
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP
,
TI
,
FS,
DE,
CL
Team Instructors
Classification Team
Team Predictor
predicts weights for every classification
Predicted weights
team in certain location
of classification teams
Location
Team Predictor:
e.g. WNN algorithm
Sample Metadata
Team Prediction:
Locality assumption
Each team has certain subdomains in the space
of instance attributes, where it is more reliable
than the others;
This assumption is supported by the experiences,
that classifiers usually work well not only in certain
points of the domain space, but in certain
subareas of the domain space [Quinlan, 1993];
If a team does not work well with the instances
near a new instance, then it is quite probable that
it will not work well with this new instance also.
Team Instructors:
Team Integrator
Constant
Team
Members
Elective
Team
Members
TC,
TM,
TP
,
TI
,
FS,
DE,
CL
Team Instructors
Classification Team
Team integrator
produces classification result for a new instance by
integrating appropriate outcomes
teams
Weightsof
of learned
classification
teams
New instance
FSi1
DEj1
CLk1
yt1
FSi2
DEj2
CLk2
yt2
FSin
DEjn
CLkn
yt1
Classification teams
Team Integrator
yt
Conclusion
Knowledge discovery with an ensemble of classifiers is known to
be more accurate than with any classifier alone [e.g. Dietterich,
1997].
If a classifier somehow consists of certain feature selection
algorithm, distance evaluation function and classification rule,
then why not to consider these parts also as ensembles making a
classifier itself more flexible?
We expect that classification teams completed from different
feature selection, distance evaluation, and classification methods
will be more accurate than any ensemble of known classifiers
alone, and we focus our research and implementation on this
assumption.
Yevgeniy Bodyanskiy
Volodymyr Kushnaryov
Existing Tools
Most existing (neuro-) fuzzy systems used for fault
diagnosis or classification are based on offline learning
with the use of genetic algorithms or modifications of
the error back propagation. When the number of features
and possible fault situations is large, tuning of the
classifying system becomes very time consuming.
Moreover, such systems perform very poorly in high
dimensions of the input space, so special modifications
of the known architectures are required.
input layer,
n inputs
output layer,
m divisors
Bodyanskiy Ye., Gorshkov Ye., Kolodyazhniy V., Wernstedt J., Probabilistic Neuro-Fuzzy Network with
Non-Conventional Activation Functions, In: Knowledge-Based Intelligent Information & Engineering
Systems, Proceedings of Seventh International Conference KES2003, 35 September, Oxford, United
Kingdom, LNAI, Springer-Verlag, 2003.
Bodyanskiy Ye., Gorshkov Ye., Kolodyazhniy V. Resource-Allocating Probabilistic Neuro-Fuzzy Network,
In: Proceedings of International Conference on Fuzzy Logic and Technology, 1012 September, Zittau,
Germany, 2003.
SIGNAL
Diagnostic out
Training
Diagnosing
12
Oracle
J ava
Program
Ontology
15
Volodymyr
Kushnaryov
Konstantin
Tatarnikov
Volodymyr
Kushnaryov
Conclusion
Industrial Ontologies Research Group (University of
Jyvaskyla), which is piloting the OntoServ.Net concept
of the Global Semantic Web - Based System for
Industrial Maintenance, has also powerful branches in
Kharkov (e.g. IOG-Kharkovs Branch, Control Systems
Research Laboratory, Data Mining Research Group,
etc.) with experts and experiences in various and
challenging data mining and knowledge discovery,
online diagnostics, forecasting and control, models
learning and integration, etc. methods, which can be and
reasonable to be successfully utilized within going-on
cooperation between Metso and Industrial Ontologies
Group.