Towards Dynamic Fuzzy Rule Interpolation with Harmony Search

Fuzzy rule interpolation enables making inferences for unmatched observations given a sparse fuzzy rule base. A sparse rule base is usually considered as a result of lacking sufficient data or human expertise to generate rules that cover the entire problem domain. Provided with a sparse rule base, an interpolating reasoning process will generate a considerable number of interpolated rules that are used to determine the outcomes of those input observations which do not match any existing rule. However, once the results are obtained, all the interpolated rules that potentially contain meaningful information about the knowledge space are discarded from the rule base. This decreases the system's overall efficiency since some of the regions covered by interpolated rules can be reused to match future observations. Thus, a dynamic fuzzy rule promotion method that picks up discarded rules and effectively adds them to the sparse rule base can improve its overall coverage of the problem space and inference efficiency. This paper proposes an initial idea to develop a dynamic fuzzy rule interpolation framework through an approach that combines the popular Transformation-based Fuzzy Rule Interpolation and Harmony Search.