Machine Learning and Qualitative Reasoning

Abstract

Qualitative modeling and reasoning is a most interesting area for applying and experimenting with machine learning techniques. Qualitative reasoning tasks of interest where machine learning can be applied include modeling, diagnosis, control, discovery, design, and knowledge compilation. This paper reviews examples of recent research into some of these applications areas. In particular, the examples given illustrate how Inductive Logic Programming (ILP; Muggleton 1990, 1992) applies naturally to these tasks when qualitative representations are used. Let us first consider the nature of descriptions that we typically encounter in qualitative reasoning. Consider the process of filling a container with water. Table 1 shows some examples of quantitative descriptions and their corresponding qualitative abstractions. In row (a) of the table, $1 and t2 denote some time points that we know exist, but the exact times are not given, z e r o and top correspond to two levels, 0 and the top of the container, where we know that z e r o < top, but the exact value of top is not known or given. In row (b) of the table, the qualitative description is read as A m o u n t is monotonical ly increasing function of L e v e l .