Generate, Test and Debug: Combining Associational Rules and Causal Models

Abstract

We present a problem solving paradigm called generate, test and debug (GTD) that combines associational rules and causal models, producing a system with both the efficiency of rules and the breadth of problem solving power of causal models. The generator uses associational rules to generate plausible hypotheses; the tester uses causal models to test the hypotheses and produce a detailed characterization of the discrepancy in case of failure. The debugger uses the ability to reason about the causal models, along with a body of domain-independent debugging knowledge, to determine how to repair the buggy hypotheses. The GTD paradigm has been implemented and tested in three different domains; we report in detail on its application to our principal domain of geologic interpretation. We also explore in some depth the character of the problems for which GTD is well suited and consider the character of the knowledge required for successful use of the paradigm.