Inductive Databases and Knowledge Scouts

Abstract

The development of very large databases and the world wide web has created extraordinary opportunities for monitoring, analyzing and predicting global economical, ecological, demographic, political, and other processes in the world. Our current technologies are, however, insufficient for these tasks, and we drowning in the deluge of data that are being collected world-wide.

New methods and integrated tools are needed that can generate goal-oriented knowledge and predictive hypotheses from massive and multimedia data, stored in large distributed databases, warehouses, and the world wide web. These methods and tools must be able to cope not only with huge data volumes in various forms, but also with data inconsistency, missing values, noise, and/or possibly weak data relevance to any given task. The development of effective methods and systems for knowledge mining in large multimedia datas emerges as a central challenge on the research agenda for the 21st century.

This talk will briefly discuss a novel project toward the above goals, which is conducted in the GMU Machine Learning and Inference Laboratory. The project concerns the development of what we call inductive databases and knowledge scouts. An inductive database extends a conventional database by integrating in it inductive inference capabilities (possibly also other types of uncertain reasoning). These capabilities allow a database to answer queries that require synthesizing plausible knowledge and make hypothetical predictions.

One of the important design conditions for an inductive database is that the hypothesized knowledge satisfy the “postulate of comprehensiblity,” that is, is in the form easy to understand and interpret by people. This can be achieved employing an appropriate representation langauge (for example, attributional calculus), and implementing a form of reasoning which we call “natural induction.” An inductive database supports the implementation of knowledge scouts, which are personal intelligent agents that “live” in the database, and automatically search for knowledge of interest to a particular user or group of users.

Presented concepts will be illustrated by initial results on searching for patterns that relate lifestyles with diseases in a large database from the American Cancer Society. At the end of the talk, we will demonstrate a system illustrating principles of natural induction.

“All human beings desire to know” Aristotle, Metaphysics, I.1.