Interactive and Opportunistic Knowledge Acquisition in Case-Based - - PowerPoint PPT Presentation
Thse soutenue publiquement pour lobtention du titre de Docteur en Informatique de lUniversit de Lyon Supervisors: A LAIN M ILLE and B ATRICE F UCHS Interactive and Opportunistic Knowledge Acquisition in Case-Based Reasoning
Thèse soutenue publiquement pour l’obtention du titre de Docteur en Informatique de l’Université de Lyon Supervisors: ALAIN MILLE and BÉATRICE FUCHS
Lyon, 13 november 2008 http://liris.cnrs.fr/amelie.cordier amelie.cordier@liris.cnrs.fr LIRIS UMR5205
Intro FIKA IAKA FRAKAS TAAABLE What’s next? 2
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
3
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
3
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
3
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
3
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
3
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
4
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Domain Knowledge Cases Similarity Knowledge Adaptation Knowledge
4
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Domain Knowledge Cases Similarity Knowledge Adaptation Knowledge
4
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Domain Knowledge Cases Similarity Knowledge Adaptation Knowledge Knowledge base
4
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
5
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Case storage is limited Knowledge acquisition is complex Knowledge acquisition is costly
6
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
7
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
8
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA IAKA FRAKAS TAAABLE
9
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Data-mining and machine learning techniques produce a lot of candidate knowledge units that are to be validated Manual acquisition is time-consuming and dependent on the expert’s availability Interactive acquisition allows the acquisition of small but valuable knowledge units
10
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Data-mining and machine learning techniques produce a lot of candidate knowledge units that are to be validated Manual acquisition is time-consuming and dependent on the expert’s availability Interactive acquisition allows the acquisition of small but valuable knowledge units
10
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Data-mining and machine learning techniques produce a lot of candidate knowledge units that are to be validated Manual acquisition is time-consuming and dependent on the expert’s availability Interactive acquisition allows the acquisition of small but valuable knowledge units
10
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA Knowledge-intensive: knowledge comes from outside the system Interactive: knowledge is acquired through interactions with the system user Opportunistic: failures trigger the acquisition process On-line: each problem-solving experience can trigger the acquisition process FIKA is a user-centric approach
11
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA Knowledge-intensive Interactive Opportunistic On-line User-centric
11
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA
CBR context Knowledge Level approach Knowledge level CBR knowledge containers Failure-driven approach
CHEF DIAL
Formalisation
Adaptation- centered approach Conservative adaptation
DÉJÀ VU
AGR
Formalisation
tation [Newell, 1982] [Richter, 1995] [Riesbeck and Schank, 1989] [Aamodt and Plaza, 1994] [Hammond, 1986] [Kinley, 2001] [Smyth and Keane, 1995] [Smyth and Keane, 1998] [Bergmann and Wilke, 1998] [Fuchs et al., 2000] [Lieber, 2007] [Leake et al., 1999] [Aamodt, 2001]
12
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA
CBR context Knowledge Level approach
Knowledge level
CBR knowledge containers Failure-driven approach CHEF DIAL Formalisation
Adaptation- centered approach Conservative adaptation DÉJÀ VU
AGR
Formalisation
tation
Systems
PADIM, [Fuchs et al., 1995],
[Fuchs, 1997]
PROLABO, [Mille et al., 1996] RADIX, [Corvaisier et al., 1997],
[Corvaisier et al., 1998]
ACCELERE, [Herbeaux, 2000],
[Herbeaux and Mille, 2007]
PIXED, [Heraud and Mille, 2000]
12
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
13
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
13
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Incremental design Evolutive design
14
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Incremental design Evolutive design
14
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA allows the acquisition of... Experiential knowledge In context Making sense Without disturbing (too much) the user
15
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA IAKA FRAKAS TAAABLE
16
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA
17
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA FRAKAS IAKA
17
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA FRAKAS IAKA IAKA-NF
17
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA FRAKAS IAKA IAKA-NF IAKA-NF(f )
17
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target ? Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target ? Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target source ? Sol(source) Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target source ? Sol(source) Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target pb2 pb1 source ? Sol(source) r3 r2 r1 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target pb2 pb1 source ? Sol(source) r3 r2 r1 Ar1 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target pb2 pb1 source ? Sol(pb1) Sol(source) r3 r2 r1 Ar1 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target pb2 pb1 source ? Sol(pb2) Sol(pb1) Sol(source) r3 r2 r1 Ar1 Ar2 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
AO target pb2 pb1 source ? Sol(pb2) Sol(pb1) Sol(source) r3 r2 r1 Ar1 Ar2 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target pb2 pb1 source ? Sol(pb2) Sol(pb1) Sol(source) r3 r2 r1 Ar1 Ar2 Ar3 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
target pb2 pb1 source Sol(target) Sol(pb2) Sol(pb1) Sol(source) r3 r2 r1 Ar1 Ar2 Ar3 Reformulation r Adaptation function Ar Adaptation operator AOr = (r, Ar) Adaptation path Adaptation method AM(source) Adaptation error associated to AOr
18
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Adaptation process
pb Sol
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Adaptation process
pb Sol Sol(source) source
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Adaptation process
pb Sol OA(source) Sol(source) source
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Adaptation process
pb Sol target OA(source) Sol(source) source
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Adaptation process
pb Sol target OA(source) OA(source) Sol(source) source g Sol(target)
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Interactions and knowledge acquisition process
pb Sol target OA(source) OA(source) Sol(source) source g Sol(target) error > ε
Role of the user
Say if the solution is correct or not Give the correct solution Give (or verify) the adaptation
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Interactions and knowledge acquisition process
pb Sol target OA(source) OA(source) OA(target) Sol(source) source Sol(target) g Sol(target) error > ε
Role of the user
Say if the solution is correct or not Give the correct solution Give (or verify) the adaptation
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Change the adaptation method?
pb Sol target OA(source) OA(source) OA(target) Sol(source) source Sol(target) g Sol(target) error > ε
Role of the user
Say if the solution is correct or not Give the correct solution Give (or verify) the adaptation
19
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Parameters studied
Size of the case base (up to 20.000 problems) Influence of the tolerance threshold Number of interactions Impact of the decomposition in steps (from 2 to 50 variables) Scope of the operators (adaptation methods) Impact of discontinuities (C∞ on Rn)
Statistical tests
Z-test Wilcoxon
20
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Modelling the domain
−10 −5 5 10−10 −5 5 10
2 z x y z
0.5 1 1.5 2 2.5
fht : R2 → R g(x, y) = sin q x2 + y2 + x 7
21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Introducing a discontinuity
−10 −5 5 10−10 −5 5 10
2 z x y z 0.2 0.4 0.6 0.8 1 1.2 1.4
fht : R2 → R h(x, y) = g(x, y) if x2 + y2 ≤ 4 g(x, y) = sin q x2 + y2 + x 7
21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Introducing a discontinuity
−10 −5 5 10−10 −5 5 10
2 z x y z
fht : R2 → R fht(x, y) = −3 − g(x, y) if x2 + y2 ≤ 4 g(x, y) = sin q x2 + y2 + x 7
21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A domain with a discontinuity
−10 −5 5 10−10 −5 5 10
2 z x y z
2 4
fht : R2 → R fht(x, y) = −3 − g(x, y) if x2 + y2 ≤ 4 g(x, y) = sin q x2 + y2 + x 7 fht(x, y) = g(x, y) if x2 + y2 ≤ 4
21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A domain with a discontinuity
−10 −5 5 10−10 −5 5 10
2 z x y z
2 4
fht : R2 → R fht(x, y) = −3 − g(x, y) if x2 + y2 ≤ 4 g(x, y) = sin q x2 + y2 + x 7 fht(x, y) = g(x, y) if x2 + y2 ≤ 4
21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Generating cases
−10 −5 5 10−10 −5 5 10
2 z x y z
2 4
fht : R2 → R fht(x, y) = −3 − g(x, y) if x2 + y2 ≤ 4 g(x, y) = sin q x2 + y2 + x 7 fht(x, y) = g(x, y) if x2 + y2 ≤ 4
21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
The system knowledge
−10 −5 5 10−10 −5 5 10
2 z x y z 21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Random generation of problems, solved by the system
−10 −5 5 10−10 −5 5 10
2 z x y z 21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Problems triggering a knowledge acquisition process...
−10 −5 5 10−10 −5 5 10
2 z x y z 21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
...are located around the discontinuity zone...
−10 −5 5 10−10 −5 5 10
2 z x y z
2 4 21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
...allowing the identification of a critical part of the domain
−10 −5 5 10−10 −5 5 10
2 z x y z
2 4 21
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
IAKA: Main contributions Formalisation of the adaptation knowledge
Adaptation operators (AOr = (r, Ar)) Adaptation methods (AM(source)) Adaptation path, adaptation step
Failure-driven knowledge acquisition strategy
Focus on adaptation knowledge acquisition Decomposition into steps to easily identify faulty knowledge
Prototype implementing the differential adaptation strategy
Numerical functions formalism Differential operators
Experiments
Test of the method with several parameters: expert tolerance threshold, number of cases, number of interactions... Experiments with CBR hypotheses, “Similar problems have similar solutions”
22
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA IAKA FRAKAS TAAABLE
23
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA
24
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA FRAKAS IAKA
24
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA FRAKAS IAKA FRAKAS-PL
24
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS properties Knowledge-intensive Interactive Opportunistic On-line User-centric FIKA FRAKAS IAKA FRAKAS-PL FRAKAS-PL(ONCO)
24
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Belief revision: updating a knowledge base while maintaining consistency Minimal change: revision operator makes a minimal change on the initial base
ψ µ ψ ◦ µ
Conservative adaptation CA◦(SDK, source ∧ Sol(source), target) = (SDK ∧ source ∧ Sol(source)) ◦ (SDK ∧ target)
25
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Belief revision: updating a knowledge base while maintaining consistency Minimal change: revision operator makes a minimal change on the initial base
ψ
ψ ◦ µ
Conservative adaptation CA◦(SDK, source ∧ Sol(source), target) = (SDK ∧ source ∧ Sol(source)) ◦ (SDK ∧ target)
25
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Belief revision: updating a knowledge base while maintaining consistency Minimal change: revision operator makes a minimal change on the initial base
CA◦(SDK, source ∧ Sol(source), target) = (SDK ∧ source ∧ Sol(source)) ◦ (SDK ∧ target)
25
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Belief revision: updating a knowledge base while maintaining consistency Minimal change: revision operator makes a minimal change on the initial base
source Sol(source) SDK target SDK target Sol(target)
Conservative adaptation CA◦(SDK, source ∧ Sol(source), target) = (SDK ∧ source ∧ Sol(source)) ◦ (SDK ∧ target)
25
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
... CBR with conservative adaptation
SDK
Analysis
result Identification
...
SDK ∧ ¬Inc
KO OK
26
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
... CBR with conservative adaptation
SDK
Analysis
result Identification
...
SDK ∧ ¬Inc
KO OK Future work: use the revision operator as a knowledge acquisition operator (SDK ◦ ¬Inc)
26
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
27
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FRAKAS- Main contributions Focus on domain knowledge
Used to perform adaptation Supposed to be consistent
One-step reasoning
Revision operator used as an adaptation operator No intermediate problems
Failure-driven knowledge acquisition strategy
Inconsistencies in the result highlight incompleteness of the available knowledge Inconsistencies are used as a basis to build additional knowledge
Prototype implementing the conservative adaptation strategy
Experiments with the conservative adaptation Use of the Dalal’s revision operator
Experiments
Applicaton domain: breast cancer treatment ...raising human-computer interaction issues
28
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
FIKA IAKA FRAKAS TAAABLE
29
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A CBR engine A cooking ontology A recipe book (the “case base”) A web interface
30
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A request in TAAABLE
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A request in TAAABLE
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A request in TAAABLE
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A failure: rhubarb is not a vegetable (from a culinary viewpoint)
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A failure: rhubarb is not a vegetable (from a culinary viewpoint)
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
A failure: rhubarb is not a vegetable (from a culinary viewpoint)
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Results after knowledge acquisition
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Results after knowledge acquisition
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Results after knowledge acquisition
31
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Incorrect knowledge Missing knowledge O request = ⇒ result Example: I asked for a sweet cake and the system offers me an olive cake recipe Example: I asked for a nut-free cake and the system offers me a peanut-butter cake U request = ⇒ result Action: correct knowledge cake = ⇒ sweet Action: add peanut-butter = ⇒ peanut O request = ⇒ result Example: I asked for a fruit pie and the system did not retrieve the rhubarb pie Example: I asked for a vegetarian dish and the system did not retrieve my favourite “omelette aux pommes de terres” U request = ⇒ result Action: add rhubarb = ⇒ fruit Action: correct knowledge eggs = ⇒ meat
More 32
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Human-computer interactions Knowledge base consistency in case of heterogeneous data sources Context through traces or case provenance ?
33
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Human-computer interactions Knowledge base consistency in case of heterogeneous data sources Context through traces or case provenance ?
33
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Human-computer interactions Knowledge base consistency in case of heterogeneous data sources Context through traces or case provenance ?
33
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Principles Models Prototypes Applications
34
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
PhD thesis document 35
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Cause of the failure Wrong knowledge (type 1 failure) Missing knowledge (type 2 failure)
A
Objective: correct faulty knowledge. Objective: complete knowledge. Example: I asked for a sweet cake and the system offers me an olive cake recipe. Example: I asked for a nut-free cake and the system offers me a peanut-butter cake. O f = ⇒ g Cause: for the system, every cake recipe is a sweet recipe. Cause: the system misses the knowledge that peanut-butter contains peanuts and thus, is not a nut-free ingredient. U f = ⇒ g Action: correct the knowledge. (cake = ⇒ sweet) Action: add a new piece of knowledge to the system, namely peanut-butter = ⇒ peanut.
B
Objective: complete knowledge. Objective: correct faulty knowledge. Example: I asked for a fruit pie and the system did not retrieve the rhubarb pie. Example: I asked for a vegetarian dish and the system did not retrieve my favourite “omelette aux pommes de terres” (and egg and potatoes dish). O f = ⇒ g Cause: the system misses the knowledge that rhubarb is a fruit. Cause: the system considers that eggs are not vegetarian ingredients. U f = ⇒ g Action: add the knowledge (rhubarb = ⇒ fruit) to the knowledge base. Action: correct the faulty knowledge (eggs = ⇒ meat).
Table: Typology of failures in TAAABLE.
36
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Aamodt, A. (2001). Modeling the knowledge contents of CBR systems. In Proceedings of the Workshop Program (at ICCBR’01), pages 32–37, Vancouver. Aamodt, A. and Plaza, E. (1994). Case-Based Reasoning: Foundational Issues, Methodological Variations, and System Approaches. AI Communications, 7(1):39–59. Bergmann, R. and Wilke, W. (1998). Towards a New Formal Model of Transformational Adaptation in Case-Based Reasoning. In Prade, H., editor, Proceedings of the 13th European Conference on Artificial Intelligence (ECAI’98), pages 53–57. John Wiley and Sons. Corvaisier, F ., Mille, A., and Pinon, J.-M. (1997). Information Retrieval on the World Wide Web using a decision making system. In Proceedings of RIAO’97, pages 284–295, Montréal. Corvaisier, F ., Mille, A., and Pinon, J.-M. (1998). RADIX 2, Assistance à la recherche d’information documentaire sur le Web. In Actes des journées d’Ingénierie des Connaissances (IC’98), pages 153–163, Pont-à-Mousson.
37
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Fuchs, B. (1997). Representation des connaissances pour le raisonnement à partir de cas. Le système ROCADE. PhD thesis, Université Jean Monnet de Saint-Etienne. Fuchs, B., Lieber, J., Mille, A., and Napoli, A. (2000). An Algorithm for Adaptation in Case-Based Reasoning. In Horn, W., editor, Proceedings of the 14th European Conference on Artificial Intelligence (ECAI’2000), pages 45–49, Berlin, Germany. IOS Press. Fuchs, B., Mille, A., and Chiron, B. (1995). Operator decision aiding by adaptation of supervision strategies . In Proceedings of the 1st International Conference on Case-Based Reasoning (ICCBR’95), pages 23–32, Sesimba, Portugal. Springer. Hammond, K. (1986). CHEF: A model of case-based planning. In Press, A., editor, Proceedings of the 5th National Conference on Artificial Intelligence, pages 267–271, Menlo Park, CA.
38
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Heraud, J.-M. and Mille, A. (2000). Pixed : vers le partage et la réutilisation d’expériences pour assister l’apprentissage. In Actes du congrès Technologies de l’information et de la communication dans les enseignements d’ingénieurs et dans l’industrie (TICE’00), pages 237–244. Herbeaux, O. (2000). ACCELERE : aide à la conception de caoutchouc cellulaire exploitant la remémoration d’expériences. PhD thesis, Université Jean Monnet, Saint Étienne. Herbeaux, O. and Mille, A. (2007). ACCELERE : système d’aide à la conception de caoutchouc cellulaire exploitant la remémoration d’expérience, volume 1 of Traité IC2 - Informatique et systèmes d’information, chapter Raisonnement à partir de cas 1 : conception et configuration de produits. Hermes Science. Kinley, A. (2001). Learning to improve case adaptation. PhD thesis, Indiana University.
39
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Leake, D., Hammond, K., Birnbaum, L., Marlow, C., and H., Y. (1999). Task-based knowledge management. In Exploring Synergies of Knowledge Management and Case-Based Reasoning, Proceedings of The American Association of Artificial Intelligence (AAAI’99) Workshop, pages 35–39, Orlando, Florida. AAAI Press. Lieber, J. (2007). Application of the Revision Theory to Adaptation in Case-Based Reasoning: the Conservative Adaptation. In Weber, R. and Richter, M., editors, Proceedings of the 7th International Conference on Case-Based Reasoning (ICCBR’07), pages 239–253, Belfast, UK. Springer-Verlag Berlin Heidelberg, LNAI 4626. Mille, A., Fuchs, B., and Herbeaux, O. (1996). A unifying framework for Adaptation in Case-Based Reasoning. In Voss, A., Bergmann, R., and Bartsch-Sporl, B., editors, Workshop on Adaptation in Case-Based Reasoning, 12th European Conference on Artificial Intelligence (ECAI’96), pages 22–28, Budapest, Hungary. Springer. Newell, A. (1982). The Knowledge Level. Artificial Intelligence, 19(2):87–127.
40
Intro FIKA IAKA FRAKAS TAAABLE What’s next?
Richter, M. M. (1995). The Knowledge Contained in Similarity Measures. Invited Talk of the First International Conference on Case-Based Reasoning (ICCBR’95). Riesbeck, C. K. and Schank, R. C. (1989). Inside Case-Based Reasoning. Lawrence Erlbaum, Hillsdale, NJ. Smyth, B. and Keane, M. T. (1995). Retrieval and Adaptation in Déjà Vu, a Case-Based Reasoning System for Software Design. In Adaptation of Knowledge for Reuse: AAAI Fall Symposium, pages 228–240. AAAI Press. Smyth, B. and Keane, M. T. (1998). Adaptation-Guided Retrieval: Questioning the Similarity Assumption in Reasoning. Artificial Intelligence, 102(2):249–293.
41