This paper describes efficient methods for exact and approximate implementation of the MIN-FEATURES bias, which prefers consistent hypotheses definable over as few features as possible. This bias is useful for learning domains where many irrelevant features are present in the training data.
We first introduce FOCUS-2, a new algorithm that...
This technical report reprints two articles that appeared in Proceedings of the Third International Machine Learning Workshop at Skytop, Pennsylvania, June 24-26, 1985. The first paper, The EG Project: Recent Progress, summarizes work on the EG project, which is investigating the role of active experimentation in aiding machine learning programs....
Test incorporations are program transformations that improve the performance of generate-and-test procedures by moving information out of the "test" and into the "generator." The test information is said to be "incorporated" into the generator so that items produced by the generator are guaranteed to satisfy the incorporated test. This article...
In recent papers on machine learning, the term 'operationalization' has been used to describe the purpose of the learning process. In particular, explanation-based learning systems are said to 'operationalize' the given target concept. Unfortunately, the exact meaning of this term has varied from one paper to another, and frequently the...
This paper formalizes a new learning from examples problem: identifying a correct concept definition from positive examples such that the concept is some specialization of a target concept defined by a domain theory. This paper describes an empirical study that evaluates three methods for solving this problem: explanation based generalization...
We introduce five criteria by which to judge the suitability of a method for solving the problem of learning concepts from examples: correctness (the correct concept should be identified), performance efficiency (the learned definition should be efficient to apply to the performance task), flexibility (the method should be able to...
The task of inductive learning from examples places constraints on the representation of training instances and concepts. These constraints are different from, and often incompatible with, the constraints placed on the representation by the performance task. This incompatibility explains why previous researchers have found it so difficult to construct good...
The performance of the error backpropagation (BP) and ID3 learning algorithms was compared on the task of mapping English text to phonemes and stresses. Under the distributed output code developed by Sejnowski and Rosenberg, it is shown that BP consistently out-performs 1D3 on this task by several percentage points. Three...
The performance of the error backpropagation (BP) and 1D3 learning algorithms was com- pared on the task of mapping English text to phonemes and stresses. Under the distributed output code developed by Sejnowski and Rosenberg, it is shown that BP consistently out- performs ID3 on this task by several percentage...