Showing papers on "Best-first search published in 1974"

A Data Base Search Problem

Abstract: The problem of constructing an efficient retrieval algorithm for responding to queries of arbitrary complexity on a relational data base is considered. A query, represented by a relation-defining expression in the relational calculus, is translated into a retrieval algorithm.

An Algorithm for the Chromatic Number of a Graph

Abstract: Christofides' algorithm for finding the chromatic number of a graph is improved both in speed and memory space by using a depth-first search rule to search for a shortest path in a reduced subgraph tree.

Some Binary Derivation Systems

Abstract: Previous studies of heuristic search techniques have usually considered situations for which the search space could be represented as a tree, which limits the applicability of the results obtained. Here Kowalski is followed and a more general formulation in terms of derivation graph is offered; the graphs correspond rather naturally to the search problems arising in automatic theorem proving and other areas. We consider a family of search procedures controlled by evaluation functions of a very general sort, having the form ƒΔ(x, Lk), where Lk is that portion of the graph generated thus far by the procedure, and the node x is a candidate for incorporation into Lk. Completeness and minimality results are obtained for a number of procedures in this family, including methods analogous to those of Moore, Dijkstra, and Pohl.

Comparison of two speech understanding systems

Abstract: This paper presents the results of a comparison of two speech systems developed at Carnegie‐Mellon University: Hear‐Say‐1 (1) and Dragon (2). Hear‐Say‐1, Which consists of cooperating but independent knowledge sources, utilizes a best first search technique of the syntatic grammer. The search terminates when the evaluation of a completed proposed sentence achieves a heuristic threshold. In this way only the most likely paths are searched, thus achieving a fast though not always accurate recognition. Dragon utilizes a combined network of grammar and phonetic spellings for a generative Markov process. This allows Dragon to search all possible paths in parallel in an amount of time that is linear with utterance length. Dragon obtains a higher accuracy of recognition but with a higher computational overhead. Both systems have been tested on the same five sets of data consisting of 102 utterances and 564 words. Hear‐Say‐1 correctly identifies about 60% of the words in about 5 to 30 times real time, depending on the number of wrong paths searched. Dragon correctly identifies about 85% of the words in an almost consistent 50 times real time. The types of errors produced by both systems are discussed. [(1) Written by D. R. Reddey, L. E. Erman, R. D. Fennell, and B. T. Lowerre, (2) Written by J. K. Baker.]

Heuristic Search and Concept Formation

Abstract: The multimodal optimum searching problem is an important essenial subproblem which is commonly contained not only in a control problem but also widely in general decision problems. This paper surveys a theoretical and experimental study of a heuristic search method of the globally optimum point of a two-dimensional, multimodal, nonlinear, and unknown criterion function by using sighted, blindfolded, and blind subjects. The main topics of the first half of the paper are basic (global) search points in the heuristic search, a shifting trend of the basic search point, omission of some trials of the basic search points, number of trials needed in the heuristic search, comparison of the search behaviors by sighted, blindfolded, and blind subjects, an extraction of heuristic search models, essential framework and appropriateness of the extracted models, deviation of number of trials due to the nationality of subjects, and so on. The second and main half of the paper studies the concept formation process in the heuristic search behavior. Concept is defined

Basic Search Patterns in Heuristic Search

Abstract: A general discussion of heuristics is followed by a discussion of a particular type of problem to be solved by a heuristic method. The problem, a common one in control theory, is the searching of the highest point on a multi-dimensional, multi-modal surface. An experiment was conducted in which human subjects performed heuristic searches to find the peak of a two-dimensional, multi-modal surface. Results of this experiment indicate a subject searches a hill according to a predetermined pattern called the basic search pattern. Using the concept of a basic search pattern a theoretical procedure for a heuristic search is postulated.