A new method of rule generation for hierarchical fuzzy systems, called a hierarchical fuzzy associative memory (HIFAM) is described. A HIFAM is structured as a binary tree and overcomes the exponential growth of the rule bases when the number of inputs increases. The training algorithm for the HIFAM is suitable for approximation and classification problems. Several benchmarks demonstrate that the proposed method compares well with existing learning techniques like artificial neural networks and decision trees.
展开▼
