Rough Set Approach to Domain Knowledge Approximation

Issue title: The 9th International Conference on Rough Sets, Fuzzy Sets, Data Mining and Granular Conputing (RSFDGrC 2003)

Article type: Research Article

Authors: Nguyen, Tuan Trung

Affiliations: Warsaw University, ul. Banacha 2, Warsaw, Poland

Abstract: Classification systems working on large feature spaces, despite extensive learning, often perform poorly on a group of atypical samples. The problem can be dealt with by incorporating domain knowledge about samples being recognized into the learning process. We present a method that allows to perform this task using a rough approximation framework. We show how human expert's domain knowledge expressed in natural language can be approximately translated by a machine learning recognition system. We present in details how the method performs on a system recognizing handwritten digits from a large digit database. Our approach is an extension of ideas developed in the rough mereology theory.

Keywords: rough mereology, concept approximation, domain knowledge, machine learning, handwritten digit recognition

Journal: Fundamenta Informaticae, vol. 59, no. 2-3, pp. 261-270, 2004