Learning to describe data in actions

Abstract

Traditional machine learning algorithms have failed to serve the needs of systems for Programming by Demonstration (PBD), which require interaction with a user (a teacher) and a task environment. We argue that traditional learning algorithms fail for two reasons: they do not cope with the ambiguous instructions that users provide in addition to examples; and their learning criterion requires only that concepts classify examples to some degree of accuracy, ignoring the other ways in which an active agent might use concepts. We show how a classic concept learning algorithm can be adapted for use in PBD by replacing the learning criterion with a set of instructional and utility criteria, and by replacing a statistical preference bias with a set of heuristics that exploit user hints and background knowledge to focus attention.