Autonomous database mining and disorder measures

Abstract

The essential question of database mining is how to computationally identify the most useful information. The discussed approach is to sift out non-useful data until useful information is discovered. This approach contrasts to existing approaches that assume they can select and rank order the most useful data when confronted by all the data, useful and non-useful. The problem with this is that it is computationally infeasible to look at all the data; so, heuristic choices are made. These choices strongly constrain what might be discovered.

Significant portions of this work was done while visiting with BISC, Computer Science University of California Berkeley, California