Learning by re-expressing concepts for efficient recognition

It is described how concepts that are accurate, though computationally inefficient for use in recognizing instances, can be re-expressed in an efficient form through a process the authors call concept operationalization.

Abstract

Much attention in the field of machine learning has been directed at the problem of inferring concept descriptions from examples. But in many learning situations, we are initrally presented with a fully-formed concept description, and our goal IS instead to re-express that description with some particular task in mind. In this paper, we specifjcally consider the task of recognizing concept instances efficiently. We describe how concepts that are accurate, though computationally inefficient for use in recognizing instances, can be re-expressed in an efficient form through a process we call concept operationalization Various techniques for concept operationalization are illustrated in the context of the LEX learning system. I