著者
Automating the Construction of Efficient Artificial , Intelligence Algorithms
タイトル
Mark W.Perlin
日時
Sep 1991
概要
The scaling up of Artificial Intelligence (AI) systems to large real-world applications requires the use of efficient underlying AI algorithms and methods. Such efficiency, however, comes at a price: the increased conceptual and implementational complexity of developing, maintaining, and refining these efficient algorithms. What is needed is a medium for rapid specification, and a mechanism for then automatically constructing efficient AI algorithms from such specifications. Recursion is an elegant and intuitive medium for algorithm specification. Standard recursive evaluation, however, may be very inefficient. Call-Graph Caching (CGC) is the preservation of the control flow of a computational process into a graph; subsequent reuse of this graph can often result in highly efficient evaluation. Our thesis is that a large number of efficient AI algorithms can be decomposed into a conceptually trasparent recursive specification, and an implementationally efficient CGC-based evaluation. This decomposition enables an AI algorithm to be rapidly specified, and then efficiently executed. In this thesis, we introduce Call-Graph Caching. We show how to use CGC to automatically construct a variety of important efficient AI algorithms, including RETE matching, Earley and Tomita parsing, linear unification, arc consistency, classical planning, and learning algorithms. We describe MatchBox and Linear MatchBox, new algorithms for incremental conjunctive match. We present CACHE, a development environment for automatically constructing and visualizing CGC-based computation. CGC evaluation transforms search into knowledge, and represents an important first step toward a unified theory of AI computation.
カテゴリ
CMUTR
Category: CMUTR
Institution: Department of Computer Science, Carnegie
        Mellon University
Abstract: The scaling up of Artificial Intelligence (AI) systems to large
        real-world applications requires the use of efficient
        underlying AI algorithms and methods.
        Such efficiency, however, comes at a price: the increased
        conceptual and implementational complexity of developing,
        maintaining, and refining these efficient algorithms.
        What is needed is a medium for rapid specification, and a
        mechanism for then automatically constructing efficient AI
        algorithms from such specifications.
        
        Recursion is an elegant and intuitive medium for algorithm 
        specification.
        Standard recursive evaluation, however, may be very inefficient.		Call-Graph Caching (CGC) is the preservation of the control
        flow of a computational process into a graph; subsequent reuse
        of this graph can often result in highly efficient evaluation.
        Our thesis is that a large number of efficient AI algorithms
        can be decomposed into a conceptually trasparent recursive
        specification, and an implementationally efficient CGC-based
        evaluation.
        This decomposition enables an AI algorithm to be rapidly
        specified, and then efficiently executed.
        
        In this thesis, we introduce Call-Graph Caching.
        We show how to use CGC to automatically construct a variety of
        important efficient AI algorithms, including RETE matching,
        Earley and Tomita parsing, linear unification, arc consistency,
        classical planning, and learning algorithms.
        We describe MatchBox and Linear MatchBox, new algorithms for 
        incremental conjunctive match.
        We present CACHE, a development environment for automatically
        constructing and visualizing CGC-based computation.
        CGC evaluation transforms search into knowledge, and represents
        an important first step toward a unified theory of AI
        computation.
Number: CMU-CS-91-176
Bibtype: TechReport
Month: Sep	
Author: Automating the Construction of Efficient Artificial 	
        Intelligence Algorithms
Title: Mark W.Perlin	
Year: 1991
Address: Pittsburgh, PA
Super: @CMUTR