In this article we approach one key aspect of the utility problem in explanation-based learning (EBL) | the expensive-rule problem |asanavoidable defect in the learning procedure. In particular, we examine the relationship between the cost of solving a problem without learning versus the cost of using a learned rule to provide the same solution, and refer to a learned rule as expensive if its use is more costly than the original problem solving from which itwas learned. The key idea we explore is that expensiveness is inadvertently and unnecessarily introduced into learned rules by the learning algorithms themselves. This becomes a particularly powerful idea when combined with an analysis tool which identi es these hidden sources of expensiveness, and modi cations of the learning algorithms which eliminate them. The result is learn-ing algorithms for which the cost of learned rules is bounded by the cost of the problem solving that they replace. We investigate this idea through an analysis of EBL Soar, an implementation of explanation-based learning within the Soar architecture. A transformational analysis is used to identify where EBL Soar inadvertently introduces substantial additional costs in the process of converting a problem solving episode into a learned rule | excessive costs which all ultimately turn out to stem from losses of information during learning. Based on these results, a modi ed EBL Soar algorithm | Bounded EBL Soar (BEBL Soar) | is developed from which all sources of expensiveness have been eliminated. The cost of using a rule learned by BEBL Soar is provably bounded by the cost of the problem solving it replaces. 1
|
540
|
Soar: An architecture for general intelligence
– Laird, Newell, et al.
- 1987
|
|
471
|
Explanationbased generalization: a unifying view
– Mitchell, Keller, et al.
- 1986
|
|
380
|
Rete: A Fast Algorithm for the Many Pattern/Many Object Pattern Match Problem
– Forgy
- 1982
|
|
331
|
Explanation-based learning: an alternative view
– Dejong, Mooney
- 1986
|
|
279
|
Quantitative results concerning the utility of explanation-based learning
– Minton
- 1988
|
|
147
|
Intelligent agents for interactive simulation environments
– TAMBE, JOHNSON, et al.
- 1995
|
|
65
|
COMPOSER: a probabilistic solution to the utility problem in speedup learning
– Gratch, DeJong
- 1992
|
|
62
|
A Preliminary Analysis of the Soar Architecture as a Basis for General Intelligence
– Rosenbloom
- 1991
|
|
50
|
The problem of expensive chunks and its solution by restricting expressiveness. Machine Learning 5(3):299{348
– Tambe, Newell, et al.
- 1990
|
|
47
|
A Statistical Approach to Solving the EBL Utility Problem
– Greiner, Jurisica
- 1992
|
|
43
|
Learning approximate control rules of high utility
– Cohen
- 1990
|
|
43
|
Machine transformation of advice into a heuristic search procedure
– Mostow
- 1983
|
|
35
|
Learning explanation-based search control rules for partial order planning
– Katukam, Kambhampati
- 1994
|
|
35
|
Ordering conjunctive queries
– Smith, Genesereth
- 1985
|
|
34
|
Finding the optimal derivation strategy in a redundant knowledge base
– Greiner
- 1991
|
|
33
|
PROLEARN: towards a prolog interpreter that learns
– Prieditis, Mostow
- 1987
|
|
32
|
Why Prodigy/EBL Works
– GRATCH, Etzioni, et al.
- 1990
|
|
31
|
Matching 100,000 learned rules
– Doorenbos
- 1993
|
|
27
|
Mapping explanation-based generalization onto soar
– Rosenbloom, Laird
- 1986
|
|
27
|
Acquiring recursive and iterative concepts with explanation-based learning
– Shavlik
- 1990
|
|
21
|
A domain independent explanation-based generalizer
– Mooney, Bennett
- 1986
|
|
17
|
The utility of EBL in recursive domain theories
– Subramanian, Feldman
- 1990
|
|
16
|
Match algorithms for generalized Rete Networks
– Lee, Schor
- 1992
|
|
16
|
Mechanization in problem solving
– Luchins
- 1942
|
|
16
|
Investigating production system representations for non-combinatorial match
– Tambe, Rosenbloom
- 1994
|
|
11
|
The Challenge of Real-time Process Control for Production Systems
– Barachini, Theuretzbacher
- 1988
|
|
11
|
Knowledge organization and the utility problem
– Doorenbos, Veloso
- 1993
|
|
11
|
Efficient matching algorithms for the SOAR/OPS5 production system
– Scales
- 1986
|
|
11
|
Soar/PSM-E: Investigating match parallelism in a learning production system
– Tambe, Kalp, et al.
- 1988
|
|
9
|
Combining left and right unlinking for matching a large number of learned rules
– Doorenbos
- 1994
|
|
9
|
A high-performance explanation-based learning algorithm
– Segre, Elkan
- 1994
|
|
8
|
Overgeneralization during knowledge compilation in soar
– Laird, Rosenbloom, et al.
- 1986
|
|
8
|
Information ltering: selection mechanisms in learning systems
– Markovitch, Scott
- 1993
|
|
7
|
Enhancementschemes for constraint processing: Backjumping, learning, and cutset decomposition
– Dechter
- 1990
|
|
6
|
Empirical and analytical performance of iterative operators
– Shell, Carbonell
- 1991
|
|
5
|
Mapping explanation-based learning onto SOAR
– Rosenbloom
- 1986
|
|
4
|
Improving example-guided unfolding
– Bostrom
- 1993
|
|
3
|
Explanation-based learning for diagnosis
– Fattah
- 1993
|
|
3
|
Real-time for RETE
– Haley
- 1987
|
|
3
|
Constraining learning with search control
– Kim, Rosenbloom
- 1993
|
|
2
|
Learning e cient rules by maintaining the explanation structure
– Rosenbloom
- 1996
|
|
2
|
Chunking in Soar: The anatomy ofa general learning mechanism
– Laird, Rosenbloom
- 1985
|
|
1
|
Learning by re-expressing concepts for e cient recognition
– Keller
- 1983
|
|
1
|
Bounding the Cost of Learned Rules: A Transformational Approach
– Kim
- 1996
|
|
1
|
On the masking e ect
– Rosenbloom
- 1993
|
