Computational Revision of Quantitative Scientiic Models

Research on the computational discovery of numeric equations has focused on constructing laws from scratch, whereas work on theory revision has emphasized qualitative knowledge. In this paper, we describe an approach to improving scientiic models that are cast as sets of equations. We review one such model for aspects of the Earth ecosystem , then recount its application to revising parameter values, intrinsic properties, and functional forms, in each case achieving reduction in error on Earth science data while retaining the communicability of the original model. After this, we consider earlier work on computational scientiic discovery and theory revision, then close with suggestions for future research on this topic. Research on computational approaches to scientiic knowledge discovery has a long history in artiicial intelligence, dating back over two decades (e.g., Lan-gley, 1979; Lenat, 1977). This body of work has led steadily to more powerful methods and, in recent years, to new discoveries deemed worth publication in the scientiic literature, as reviewed by Langley (1998). However, despite this progress, mainstream work on the topic retains some important limitations. One drawback is that few approaches to the intelligent analysis of scientiic data can use available knowledge about the domain to constrain search for laws or explanations. Moreover, although early work on computational discovery cast discovered knowledge in notations familiar to scientists, more recent eeorts have not. Rather, innuenced by the success of machine learning and data mining, many researchers have adopted formalisms developed by these elds, such as decision trees and Bayesian networks. A return to methods that operate on established scientiic notations seems necessary for scientists to understand their results.