Integrating Many Techniques for Discovering Structure in Data

This paper describes a formal representation of the discovery process that integrates of any number of data analysis strategies, regardless of their differences. We have implemented a system based on this formalization, called the Scientist's Empirical Assistant (SEA). SEA employs several analysis strategies from the discovery literature, including techniques for function finding, causal modeling, and Bayesian conditioning. It uses high-level knowledge about the discovery process, the strategies, and the domain of study to coordinate the selection and application of analyses. It relies on the skills and initiatives of an expert user to guide its search for structure. Finally, it designs and runs experiments with a simulator to verify its findings. SEA's primary sources of power are its abstraction of the discovery process and its numerous analysis strategies.