Multi-dimensional encoding to reduce bias in fuzzy knowledge-bases

This paper presents preliminary study into dimensional bias introduced when a naturaly multi-dimensional structure (eg. a fuzzy logic knowledge-base) is encoded as a single-dimensional string for use in an evolutionary algorithm. The evolutionary algorithm is modified to use a multi-dimensional encoding in favour over a single-dimensional string, preserving the multi-dimensional nature of a fully-specified fuzzy logic knowledge-base. Experiments show a clear benefit in function and data approximation applications, but show no differene in the chosen control application. Side benefits to using multi-dimensional encoding are foud that make it worth considering even if a particular problem does not suffer from significant dimensional bias.