Learning from Imperfect Data

Reference: P. Brazdil and P. Clark. Learning from Imperfect Data. In P. B. Brazdil and K. Konolige, editors, Machine Learning, Meta-reasoning and Logics, pages 207-232, Boston, 1990. Kluwer.

Abstract: Systems interacting with real-world data must address the issues raised by the possible presence of errors in the observations it makes. In this paper we first present a framework for discussing imperfect data and the resulting problems it may cause. We distinguish between two categories of errors in data -- random errors or `noise', and systematic errors -- and examine their relationship to the task of describing observations in a way which is also useful for helping in future problem-solving and learning tasks. Secondly we proceed to examine some of the techniques currently used in AI research for recognising such errors.