Data modeling: a process for pattern induction
Marsha J. Meredith · Journal of Experimental & Theoretical Artificial Intelligence · 1991
The ability to perceive and manipulate patterns is a fundamental part of human intelligence in general, and of scientific discovery in particular. In scientific discovery a pattern-characterizing concept must be induced from data, tested against subsequent data, and reformulated until it explains the data observed. This research develops and tests one approach to modeling the process of pattern induction and reformulation as data continually become available. In this approach each new piece of data contributes to the construction of a data model, a description of the observed data that organizes or systematizes them. The data model serves as a bridge between the raw data and the world of pattern concepts, eventually suggesting and providing evidence in support of an explanatory conceptual model of the pattern. The data modeling approach is applied to the domain of integer sequence patterns, a classic inductive domain. Strengths and weaknesses of the approach and comparisons to human performance are made.