12-Case Based Reasoning(Jun Yin).

advertisement
Knowledge Learning by Using Case Based Reasoning (CBR)
Knowledge Learning by Using Case Based
Reasoning (CBR)
Jun Yin and Yan Meng
Department of Electrical and Computer Engineering
Stevens Institute of Technology
Hoboken, NJ, USA
4/9/2015
1
Knowledge Learning by Using Case Based Reasoning (CBR)
What’s CBR?
• Case-Based Reasoning (CBR) is a name given to a reasoning
method that solves a new problem by remembering a previous
similar experiences and by reusing information and knowledge
of that situation.
• Ex: Medicine
– doctor remembers previous patients especially for rare combinations of
symptoms
• Ex: Law
– English/US law depends on precedence
– case histories are consulted
4/9/2015
2
Knowledge Learning by Using Case Based Reasoning (CBR)
CBR System Components
• Case-base
– database of previous cases (experience)
• Retrieval of relevant cases
– matching most similar case(s)
– retrieving the solution(s) from these case(s)
• Adaptation of solution
– alter the retrieved solution(s) to reflect differences between
new case and retrieved case(s)
4/9/2015
3
Knowledge Learning by Using Case Based Reasoning (CBR)
The Case Based Reasoning Cycle
Problem
SIMILAR CASES
New
case
RETAIN
RETRIEVE
PRIOR
CASES
R
E
U
S
E
CASE-BASE
Solution
REVISE
Solution
Knowledge Learning by Using Case Based Reasoning (CBR)
Case Retrieval and Adaptation
• Case retrieval
– the process of finding within the case base those cases that
are the closest to the current case.




Nearest Neighbor Retrieval
Inductive approaches
Knowledge Guided Approaches
Validated Retrieval
• Case Adaptation
– the process of translating the retrieved solution into the
solution appropriate for the current problem.
4/9/2015
5
Knowledge Learning by Using Case Based Reasoning (CBR)
Open Tools
•
freeCBR
 is a free open source Java implementation of a "Case
Based Reasoning" engine. (http://freecbr.sourceforge.net/)
•
myCBR
 is an open-source case-based reasoning tool developed
at DFKI. (http://mycbr-project.net/index.html)
4/9/2015
6
Knowledge Learning by Using Case Based Reasoning (CBR)
freeCBR
a very small case set:
4/9/2015
7
Knowledge Learning by Using Case Based Reasoning (CBR)
freeCBR (cont.)
search from the case set:
the result of the search:
4/9/2015
8
Knowledge Learning by Using Case Based Reasoning (CBR)
Open Tool – myCBR
4/9/2015
9
Knowledge Learning by Using Case Based Reasoning (CBR)
Open Tools – freeCBR & myCBR
Modeling Similarity Measures:
These two tools follow the approach in which, for an attribute-value based
case representation consisting of n attributes, the similarity between a query q
and a case c may be calculated as follows:
n
Sim(q, c)   i  simi (qi , ci )
i 1
Here, simi and wi denote the local similarity measure and the weight of
attribute i, and Sim represents the global similarity measure.
4/9/2015
10
Knowledge Learning by Using Case Based Reasoning (CBR)
Case Retrieval
• Nearest Neighbor Retrieval
 Retrieve most similar
 k-nearest neighbor
- k-NN
- like scoring in bowls or curling
 Example
- 1-NN
- 5-NN
4/9/2015
11
Knowledge Learning by Using Case Based Reasoning (CBR)
Case Retrieval
• Decision Tree
 e.g.
Case-Base indexed
using a decision-tree
4/9/2015
12
Knowledge Learning by Using Case Based Reasoning (CBR)
Case Retrieval
• We propose a self-organizing reservoir computing based
network for case retrieval.
Case
Base
Query
Previous Cases
q
Case Retriecal
The Most
Similar Case
y(t)
Self-organizing RC
based network
x(t)
Self-Organizing
Topology with SNN
Self-organizing Reservoir Computing
based Network architecture
4/9/2015
13
,
Knowledge Learning by Using Case Based Reasoning (CBR)
Case Retrieval
• Benchmark to evaluate the performance of proposed RC based
network.
 NARMA task
- The Nonlinear Auto-Regressive Moving Average
(NARMA) task consists of modeling the output of the
following tenth-order system :
y (t  1)  0.3 y (t )  0.05 y (t )[ i 0 y (t  i)]  1.5u (t  9)u (t )  0.1
9
4/9/2015
14
Knowledge Learning by Using Case Based Reasoning (CBR)
NARMA task:
#1
0.6
#2
expected values
estimated values
expected values
estimated values
0.6
0.5
0.5
0.4
values
values
0.4
0.3
0.2
0.2
0.1
0
0.3
0.1
10
20
30
40
50
number
60
70
80
90
100
0
10
20
30
40
50
number
60
70
80
90
100
Mean squared error = 0.128221, std = 0.0200301
4/9/2015
15
Knowledge Learning by Using Case Based Reasoning (CBR)
Future Work
•
Integrate RC based network into CBR system
• Develop the CBR system based on existing tools for more
complicated tasks
4/9/2015
16
Download