When a fuzzy input vector is presented to a multi-layerfeedforward neural network, the corresponding fuzzy output vector iscalculated by fuzzy arithmetic. It is well known that fuzzy arithmeticinvolves excess fuzziness; we employ subdivision methods of intervalinput vectors for decreasing excess fuzziness included in fuzzy outputsfrom neural networks. First we examine a simple subdivision method whereeach level set of a fuzzy input vector is subdivided into many cellswith the same size by uniformly subdividing all elements of the levelset into multiple intervals. Next we examine a hierarchical subdivisionmethod where each level set is subdivided into many cells with differentsizes by iteratively subdividing a single element of a cell into twointervals. Finally we modify the hierarchical subdivision method forefficiently decreasing excess fuzziness
展开▼