Many learning systems search through a space of possible performance elements, seeking an element whose expected utility, over the distribution of problems, is high. As the task of finding the globally optimal element is often intractable, many practical learning systems instead hill-climb to a local optimum. Unfortunately, even this is problematic as the learner typically does not know the underlying distribution of problems, which it needs to determine an element's expected utility. This paper addresses the task of approximating this hill-climbing search when the utility function can only be estimated by sampling. We present a general algorithm, palo, that returns an element that is, with provably high probability, essentially a local optimum. We then demonstrate the generality of this algorithm by presenting three distinct applications, that respectively find an element whose efficiency, accuracy or completeness is nearly optimal. These results suggest approaches to solving the utility problem from explanation-based learning, the multiple extension problem from nonmonotonic reasoning and the tractability/completeness tradeoff problem from knowledge representation. 1
|
7709
|
Computers and Intractability: A Guide to the Theory of NP-Completeness
– Garey, Johnson
- 1979
|
|
3214
|
C4.5: Programs for Machine Learning
– Quinlan
- 1993
|
|
2172
|
Optimization by simulated annealing
– Kirkpatrick, Gelatt, et al.
- 1983
|
|
1364
|
A theory of the learnable
– Valiant
- 1984
|
|
1154
|
Random graphs
– Bollobás
- 1985
|
|
715
|
Probability inequalities for sums of bounded random variables
– Hoeffding
- 1963
|
|
637
|
Estinwtion of Dependences Based on Empirical Data
– Vapnik
- 1982
|
|
540
|
Soar: An architecture for general intelligence
– Laird, Newell, et al.
- 1987
|
|
515
|
Proof verification and hardness of approximation problems
– Arora, Lund, et al.
- 1992
|
|
496
|
The complexity of theorem-proving procedures
– Cook
- 1971
|
|
490
|
Irrelevant features and the subset selection problem
– John, Kohavi
- 1994
|
|
481
|
An analysis of time-dependent planning
– Dean, Boddy
- 1988
|
|
427
|
A measure of asymptotic efficiency for tests of a hypothesis based on the sum of observations
– Chernoff
- 1995
|
|
322
|
Decision theoretic generalization of the PAC model for neural net and other learning applications
– Haussler
- 1992
|
|
273
|
Connectionist learning procedures
– Hinton
- 1989
|
|
251
|
Learning in Embedded Systems
– Kaelbling
- 1993
|
|
219
|
Quantifying Inductive Bias: AI Learning Algorithms and Valiant’s Learning Framework
– Haussler
- 1988
|
|
210
|
Solving TimeDependent Planning Problems
– Boddy, Dean
- 1989
|
|
166
|
Preferred subtheories: an extended logical framework for default reasoning
– Brewka
- 1989
|
|
165
|
Greedy Attribute Selection
– Caruana, Freitag
- 1994
|
|
162
|
Logical Foundation of Artificial Intelligence
– Genesereth, Nilsson
- 1987
|
|
157
|
Classifier systems and genetic algorithms
– Booker, Goldberg, et al.
- 1989
|
|
149
|
Explanation-based learning: A problem solving perspective
– Minton, Carbonell, et al.
- 1989
|
|
133
|
Default reasoning, nonmonotonic logics and frame problem
– Hanks, McDermott
- 1986
|
|
120
|
Theorist: A logical reasoning system for defaults and diagnosis
– Poole, Goebel, et al.
- 1987
|
|
116
|
Theory refinement combining analytical and empirical methods
– Ourston, Mooney
- 1994
|
|
113
|
Shift of bias for inductive concept-learning
– Utgoff
- 1986
|
|
111
|
Bandit Problems: Sequential Allocation of Experiments
– Berry, Fristedt
- 1985
|
|
110
|
Two Theses of Knowledge Representation: Language Restrictions, Taxonomic Classification, and The Utility of Representation Services
– Doyle, Patil
- 1991
|
|
109
|
Learning Search Control Knowledge: An Explanation–Based Approach
– Minton
- 1988
|
|
105
|
Learning Automata --- an Introduction
– Narendra, Thathachar
- 1989
|
|
93
|
Making believers out of computers
– Levesque
- 1986
|
|
93
|
Knowledge compilation using Horn approximation
– Sellman, Kautz
- 1991
|
|
77
|
Linear time algorithms for testing the satisfiability of propositional horn formulae
– Dowling, Gallier
- 1984
|
|
72
|
LEAP: A learning apprentice for VLSI design
– Mitchell, Mahadevan, et al.
- 1985
|
|
69
|
Nonmonotonic reasoning
– Reiter
- 1987
|
|
65
|
COMPOSER: a probabilistic solution to the utility problem in speedup learning
– Gratch, DeJong
- 1992
|
|
61
|
Hoeffding Races: Accelerating Model Selection Search for Classification and Function Approximation
– Maron, Moore
- 1994
|
|
60
|
On the declarative semantics of stratified deductive databases and logic programs
– Przymusinski
- 1988
|
|
52
|
Optimal problemsolving search: All-or-none solutions
– Simon, Kadane
- 1975
|
|
47
|
A Statistical Approach to Solving the EBL Utility Problem
– Greiner, Jurisica
- 1992
|
|
39
|
Provably bounded optimal agents
– Russell, Subramanian, et al.
- 1993
|
|
36
|
Twenty-seven principles of rationality
– Good
- 1971
|
|
34
|
Finding the optimal derivation strategy in a redundant knowledge base
– Greiner
- 1991
|
|
34
|
Defining operationality for explanation-based learning
– Keller
- 1988
|
|
33
|
The reference class
– Kyburg
- 1983
|
|
29
|
Hierarchical knowledge bases and efficient disjunctive reasoning
– Borgida, Etherington
- 1989
|
|
29
|
Vivid knowledge and tractable reasoning: Preliminary report
– Etherington, Borgida, et al.
- 1989
|
|
29
|
Learning to select useful landmarks
– Greiner, Isukapalli
- 1996
|
|
28
|
Generalizing prioritization
– Grosof
- 1991
|
|
