Publications

"A Framework for Autonomous Knowledge Discovery from Databases"

by Gary Ray Livingston, University of Pittsburgh, May, 2001

WWW Link: www.cs.pitt.edu/~gary/dissertation.ps

Author's email: gary@cs.pitt.edu

Abstract:

This research evaluates the sufficiency of an agenda- and justification-based framework for fully autonomous discovery systems. The proposed framework provides a reasoning component for the autonomous selection of discovery tasks which ranks the tasks by their plausibility. A task's plausibility is computed from the interestingness of the items involved in the tasks and the strengths of justifications given for performing them. Heuristics are used to perform tasks and to propose new tasks. In addition, the framework is extremely modular, facilitating the extension of a discovery system.

The framework's sufficiency was demonstrated by implementing it in a prototype system called HAMB and using it to make discoveries from the domain of experimental conditions that favor the growth of crystals of DNA-protein complexes and proteins for X-ray crystallographic studies. Details of the prototype's implementation, such as its interestingness function and its heuristics for performing, proposing, and justifying tasks, are reviewed, and results of evaluations of the framework and HAMB are presented and discussed.

Empirical results suggest that reasons given for performing tasks and estimates of the interestingness of items involved in the tasks are useful in guiding the discovery process and that implementing intelligent datamining assistants and fully autonomous discovery agents may be feasible.