FLARE: Induction with Prior Knowledge
Christophe G. Giraud-Carrier · 1996
This paper discusses a general framework called FLARE, that integrates inductive learning using prior knowledge together with reasoning in a non-recursive, propositional setting. FLARE learns incrementally by continually revising its knowledge base in the light of new evidence. Prior knowledge is generally given by a teacher and takes the form of pre-encoded rules. Simple defaults, combined with similarity-based reasoning and learning capabilities, enable FLARE to exhibit reasoning that is normally considered non-monotonic. The framework is particularly useful in the context of knowledge acquisition and discovery, as theory and experience are combined. Results of several experiments are reported to demonstrate FLARE's applicability. 1. INTRODUCTION One of the greatest challenges in the construction of intelligent or expert systems is knowledge acquisition. Traditionally, knowledge acquisition consists of extracting domain knowledge from human experts (via interviews, etc.) and careful...