Learning 10, 000 Chunks: What's It Like Out There?

This paper describes an initial exploration into large learning systems, i.e., systems that learn a large number of rules. Given the well-known utility problem in learning systems, efficiency questions are a major concern. But the questions are much broader than just efficiency, e.g., will the effectiveness of the learned rules change with scale? This investigation uses a single problem-solving and learning system, Dispatcher-Soar, to begin to get answers to these questions. Dispatcher-Soar has currently learned 10, 112 new productions, on top of an initial system of 1, 819 productions, so its total size is 11, 931 productions. This represents one of the largest production systems in existence, and by far the largest number of rules ever learned by an AI system. This paper presents a variety of data from our experiments with Dispatcher-Soar and raises important questions for large learning systems.

[1]  Charles L. Forgy,et al.  Rete: A Fast Algorithm for the Many Patterns/Many Objects Match Problem , 1982, Artif. Intell..

[2]  Allen Newell,et al.  R1-Soar: An Experiment in Knowledge-Intensive Programming in a Problem-Solving Architecture , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[3]  Nancy Martin,et al.  Programming Expert Systems in OPS5 - An Introduction to Rule-Based Programming(1) , 1985, Int. CMG Conference.

[4]  John E. Laird,et al.  Mapping Explanation-Based Generalization onto Soar , 1986, AAAI.

[5]  Daniel P. Miranker TREAT: a better match algorithm for AI production systems , 1987, AAAI 1987.

[6]  Kemal Oflazer,et al.  Partitioning in parallel processing of production systems , 1987 .

[7]  Anoop Gupta Parallelism in production systems , 1987 .

[8]  Allen Newell,et al.  SOAR: An Architecture for General Intelligence , 1987, Artif. Intell..

[9]  Jaime G. Carbonell,et al.  Learning effective search control knowledge: an explanation-based approach , 1988 .

[10]  Anoop Gupta,et al.  Comparison of the RETE and TREAT production matchers for soar (A summary) , 1988, AAAI 1988.

[11]  Virginia E. Barker,et al.  Expert systems for configuration at Digital: XCON and beyond , 1989, Commun. ACM.

[12]  Paul S. Rosenbloom,et al.  Knowledge Level and Inductive Uses of Chunking (EBL) , 1990, AAAI.

[13]  William W. Cohen Learning Approximate Control Rules of High Utility , 1990, ML.

[14]  A. Newell Unified Theories of Cognition , 1990 .

[15]  Steven Minton,et al.  Quantitative Results Concerning the Utility of Explanation-based Learning , 1988, Artif. Intell..

[16]  Allen Newell,et al.  A Preliminary Analysis of the Soar Architecture as a Basis for General Intelligence , 1991, Artif. Intell..

[17]  Milind Tambe Eliminating combinatorics from production match , 1991 .