SLIDES FROM "Twenty-Seventh International Joint Conference on Artificial Intelligence IJCAI 2018"
Case-Based Reasoning (CBR) learns new knowledge from data and so can cope with changing environments. CBR is very different from model-based systems since it can learn incrementally as new data is available, storing new cases in its case-base. This means that it can benefit from readily available new data, but also case-base maintenance (CBM) is essential to manage the cases, deleting and compacting the case-base. In the 50th anniversary of CNN (considered the first CBM algorithm), new CBM methods are proposed to deal with the new requirements of Big Data scenarios. In this paper, we present an accessible historic perspective of CBM and we classify and analyse the most recent approaches to deal with these requirements.
Boost Fertility New Invention Ups Success Rates.pdf
Maintenance of Case Bases: Current Algorithms after Fifty Years. J.M. Juarez, S. Craw et al. IJCAI 2018
1. J. M. Juarez1, S. Craw2, J. Ricardo Lopez-Delgado1 , M. Campos1
1 University of Murcia
2 Robert Gordon University
h"ps://www.ijcai.org/proceedings/2018/0770.pdf
Maintenance of case bases:
current algorithms after fifty years
July
16
IJCAI-‐ECAI
2018
,
Stockholm
h"p://www.ijcai-‐18.org/
FUNDED BY MINECO WITH ID
PROJECT TIN2013-45491-R.
2. 2
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
50years
3. 3
Picture from:
https://sites.google.com/a/peterhart.net/peterhart/Home
“The Condensed Nearest Neighbor Rule," IEEE Trans. on Information
Theory, Vol. IT-14, No. 3, pp 515-516 (May 1968)
Peter E. Hart
CNN algorithm
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
4. 4
“The Condensed Nearest Neighbor Rule," IEEE Trans. on Information
Theory, Vol. IT-14, No. 3, pp 515-516 (May 1968)
z
eb
d
c
xy
CNN Algorithm
Input: original case-base C
Output: maintained case-base CM
CMßempty
cß first case of C
Cß C – {c}
CMß CM U {c}
Repeat
For all c in C
Cß C – {c}
if not CorrecClassifyNN(c,CM)
CMß CM U {c}
endif
enfor
Until C without changes
Return CM
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
5. • OUTLINE:
1. 50
years
map
of
CBM
algorithms
2. Advances
last
5
years
3. Conclusions
5
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
6. • CASE-‐BASED
REASONING
(CBR)
– Solves
by
retrieving
similar
problems
(cases)
• Retrieve,
Adapt,
Learn
–
≠
model-‐based
systems
• No
structure:
atomic
case
=
(problem,
solu[on)
• Case
Base:
Knowledge
incremented
dynamically
– Challenge:
number
of
cases
– Approach:
Case-‐Base
Maintenance
(CBM)
alg.
6
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
7. 7
NN period
1997-2000
DROP 1 2 3 4 5
Wilson-Martinez
1995
Competence Def.
Smyth-Keane
1968
CNN
Hart
1972
RNN
Gates
1972
ENN
Gates
1975
SNN
Ritter et al
1976
RENN
Tomek
1976
All-KNN
Tomek
1987
SHRINK
Kibler-Aha
1991
IB1 IB2 IB3 IB4
Aha-Kibler
1997
Introsp. Learn
Bonzano et al
• 50
years
CBM
map
B.
Smyth
and
E.
McKenna.
Competence
models
and
the
maintenance
problem.
Computa[onal
Intelligence,
17(2):
235-‐249,
2001.
!=
Machine
Learning:
purpose
of
each
case
in
CBR
COMPETENCE:
capacity
of
each
case
to
solve
- Coverage
- Reachability
- Competence
Groups
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
8. 8
NN period
Competence period
1997-2000
DROP 1 2 3 4 5
Wilson-Martinez
1995
Competence Def.
Smyth-Keane
1968
CNN
Hart
1972
RNN
Gates
1972
ENN
Gates
1975
SNN
Ritter et al
1976
RENN
Tomek
1976
All-KNN
Tomek
1987
SHRINK
Kibler-Aha
1991
IB1 IB2 IB3 IB4
Aha-Kibler
1997
Introsp. Learn
Bonzano et al
• 50
years
CBM
map
Meta-CB
CumminsDerek
2011
IBL-DS
Beringer-Hüllermeier
2007
CTE
Craw et al
2007
CBE
Delany-Cunningham
2004
COV RFD RC
McKenna-Smyth
2001-2000
Maint. Retrieval
Craw-Jarmulak
2001
KeepItSimple
Yang-Wu
2000
ICF
Brighton-Mellis
1999
Competence Model
Smyth-McKenna
1998
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
9. • OUTLINE:
1. 50
years
map
of
CBM
algorithms
2. Advances
last
5
years
3. Conclusions
9
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
10. • Last
5
years
– Intense
research
– Novel
(theore[cal)
models
– New
problems
(non
exis[ng)
– Real
life
solu[ons
10
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
11. 11
NN period
Competence period
2014
Closure
Lu et al
2014
WCOID-GM
Smiti-Elouedi
2014
T-CBM
Lupiani etal
2015
CM
Chebel et al
2015
FFD
Leake-Schack
2015
Pref-CBM
Abdel-Hüllermeier
2015
CBNI
Yamamoto et al
2016
NEFCS/SSR
Lu etal
2016
MOE-CBM
Lupiani etal
2017
FP-CA
Mathew-Chakraborti
2017
SCBM
Smiti-Elouedi
Competence Improvements Re-structuring case-bases Time Complexity
1997-2000
DROP 1 2 3 4 5
Wilson-Martinez
1995
Competence Def.
Smyth-Keane
1968
CNN
Hart
1972
RNN
Gates
1972
ENN
Gates
1975
SNN
Ritter et al
1976
RENN
Tomek
1976
All-KNN
Tomek
1987
SHRINK
Kibler-Aha
1991
IB1 IB2 IB3 IB4
Aha-Kibler
1997
Introsp. Learn
Bonzano et al
Meta-CB
CumminsDerek
2011
IBL-DS
Beringer-Hüllermeier
2007
CTE
Craw et al
2007
CBE
Delany-Cunningham
2004
COV RFD RC
McKenna-Smyth
2001-2000
Maint. Retrieval
Craw-Jarmulak
2001
KeepItSimple
Yang-Wu
2000
ICF
Brighton-Mellis
1999
Competence Model
Smyth-McKenna
1998
• Last
5
years:
CBM
map
Period 2014-2018
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
12. • Last
5
years:
complex
problems
– MulM-‐objecMve
opMmizaMon
• Search
Space:
Case-‐Base
• Size
vs.
Accuracy
• Noise
vs.
Redundancy
• MOEA:
NSGA-‐II
• OpMmizaMon:
– Min(no
redundant
cases)
– Min(dist.
Non-‐redundant)
– Max(accuracy
system)
12
E.Lupiani, S.Massie, S.Craw, J.M. Juarez, and J. Palma. Case-base
maintenance with multi-objective evolutionary algorithms. Journal of Intelligent
Information Systems, 46(2):259–284, Apr 2016..
MOE-CBM Lupiani-Massie-Craw et al 2016
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
13. • Last
5
years:
improving
competence
– Closure-‐Competence
Model
• Inadequate
Comp.
Model
• Disjoint
parMMons
(consistent)
13
z
e
b
d
c
xy
Competence closure and Related closure exampleNing Lu, Guangquan Zhang, and Jie Lu. Concept drift detection via competence
models. Artificial Intelligence, 209:11–28, April 2014.
z
eb
d
c
xy
CovSet(c)
ReachSet(c)
RelSet(e)
Competence
Group
CompetenceGroup
ORIGINAL COMPETENCE MODEL SMYTH-MCKENNA 2011
Lu-Zhang-Lu Competence Model Proposal 2014
CompetenceClosure
RelSet(y)
RelSet(z)
RelSet(x)
RelClosure(c)=RelSet(x) U RelSet(y) U RelSet(z)
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
14. • Last
5
years:
temporal
dimension
(real
life
problems)
– Temporal
CBM
• Case:
sequence
of
events
• Adapt
CBM
approaches
14
Monitoring Elderly People At Home: falling detection
E. Lupiani, J.M. Juarez, J. Palma, and Roque Marin. Monitoring elderly people
at home with temporal case-based reasoning. Knowledge-Based Systems,
134:116 – 134, 2017.
E.Lupiani, J.M. Juarez, and J. Palma. A proposal of temporal case-base
maintenance algorithms. In Luc Lamontagne and Enric Plaza, editors, Case-
Based Reasoning Research and Development, pages 260–273. Springer, 2014.
T-CBM Lupiani-Juarez-Palma 2014
z
eb
d
c
xy
TEMPORAL MAINTENANCE: T-CNN
T-RENN
T-DROP1-3
T-ICF
Etc.
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
15. • Last
5
years:
re-‐structuring
case-‐bases
– Flexible
Feature
DeleMon
• Remove
part
case
• Less
competence
loss
• High
dimensional
problems
15
Example 4-feature case
D. Leake and B. Schack.Flexible feature deletion: Compacting case bases by
selectively compressing case contents. In Eyke Hüllermeier and Mirjam Minor,
editors, Case-Based Reasoning Research and Development, pages 212–227.
Springer, 2015.
FFD Leak-Schack, 2015
Orthogonal Deletion 2 features
3-atrribute
case
Local Feature Deletion
4-atrribute
case
3-atrribute
case
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
16. • Last
5
years:
re-‐structuring
case-‐bases
– Preference
CBR
Model
• SoluMon
Si
preferred
to
Sj
• Decompose
case
structure
• c=(prob,sol,PrefSet)
16
Example Pref-CBM approach
A. Abdel-Aziz and E. Hüllermeier. Case base maintenance in preference based
CBR. In Eyke Hüllermeier and Mirjam Minor, editors, Case-Based Reasoning
Research and Development, pages 1–14. Springer, 2015..
Pref-CBR approach Abdel-Aziz & Hüllermeier, 2015
SOLUTION DIMENSION (PREFERENCES)
distance
z
eb
d
c
xy
PROBLEM DIMENSION:
s
s
s s
s
s
sc se
NN(c)={e}
preferred
preferred preferred
preferred
preferred
preferred
preferred
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
17. • OUTLINE:
1. 50
years
map
of
CBM
algorithms
2. Advances
last
5
years
3. Conclusions
17
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
18. • Conclusions
– Deep
Impact
of
Competence
Model
Smyth-‐McKenna-‐Keane
– New
potenMal
CBR
applicaMons
(not
explored
in
depth):
• Interpretable
reducMon
of
high
dimensional
problems:
e.g.
Flexible
Feature
DeleMon[Leake&Schack,2015]
• Social
network
datasets:
e.g.
ComposiMonal
AdaptaMon
[Mathew&ChakraborM,2017]
– ApplicaMons
in
the
Industry:
• Monitoring:
T-‐CBM
[Lupiani
et
al
2015]
• Long
term
use
of
intelligent
systems:
Dria-‐CBM
[Lu
et
al
2016]
– Future
direcMons:
• Fair
comparisons
(now
limited
to
classical
CBM
alg.)
18
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
19. 19
A. Abdel-Aziz
D. Aha
M.K. Albert
J. Beringer
A. Bonzano
D. Bridge
H. Brighton
S. Chakraborti
B. Chebel-Morello
S. Craw
L. Cummins
P. Cunningham
Z. Elouedi
G. Gates
E. Golobardes
P. Hart
E. Hüllermeier
T.L. Isenhour
S. J. Delany
J. Jarmulak
J.M. Juarez
M. K. Haouchine
T. Kawabe
M.T. Keane
D. Kibler
R. Knauf
Y. Kobayashi
D. Leake
R. Lopez de Mantaras
S.R. Lowry
J. Lu
N. Lu
E. Lupiani
R. Marin
T.R. Martinez
S. Massie
D. Mathew
E. McKenna
C. Mellish
J. Palma
G.L. Ritter
R. Rowe
M. Salamó
B. Schack
A. Smiti
B. Smyth
Y. Sukarai
I. Tomek
S. Tsuruta
D.R. Wilson
N. Wiratunga
H.B. Woodruff
J. Wu
Y. Yamamoto
Q. Yang
N. Zerhouni
G. Zhang
Acknowledgements
(CBM
algorithms
authors)
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018
20. 20
A. Abdel-Aziz
D. Aha
M.K. Albert
J. Beringer
A. Bonzano
D. Bridge
H. Brighton
S. Chakraborti
B. Chebel-Morello
S. Craw
L. Cummins
P. Cunningham
Z. Elouedi
G. Gates
E. Golobardes
P. Hart
E. Hüllermeier
T.L. Isenhour
S. J. Delany
J. Jarmulak
J.M. Juarez
M. K. Haouchine
T. Kawabe
M.T. Keane
D. Kibler
R. Knauf
Y. Kobayashi
D. Leake
R. Lopez de Mantaras
S.R. Lowry
J. Lu
N. Lu
E. Lupiani
R. Marin
T.R. Martinez
S. Massie
D. Mathew
E. McKenna
C. Mellish
J. Palma
G.L. Ritter
R. Rowe
M. Salamó
B. Schack
A. Smiti
B. Smyth
Y. Sukarai
I. Tomek
S. Tsuruta
D.R. Wilson
N. Wiratunga
H.B. Woodruff
J. Wu
Y. Yamamoto
Q. Yang
N. Zerhouni
G. Zhang
Acknowledgements
(CBM
algorithms
authors)
THANK
YOU
FOR
YOUR
ATTENTION
Jose
M.
Juarez
jmjuarez@um.es
DOWNLOAD
PAPER
(OPEN
ACCESS)
:
h"ps://www.ijcai.org/proceedings/2018/0770.pdf
J.
M.
Juarez,
S.
Craw,
J.
Ricardo
Lopez-‐Delgado,
M.
Campos.
Maintenance
of
case
bases:
current
algorithms
aOer
fiOy
years.
IJCAI
2018