Towards a high capacity fuzzy associative memory model
Fu-Lai Chung, Tong Lee · 2002
Kosko's fuzzy associative memory (FAM) is the very first example to use neural networks to articulate fuzzy rules for fuzzy systems. Despite its simplicity and modularity, the model suffers from extremely low memory capacity, i.e., single rule pattern storage, and hence it is limited to small rule-base applications. In this paper, a high capacity FAM model called fuzzy relational memory (FRM) is proposed. Based upon the well-developed theoretical results of solving fuzzy relational equations, a theorem for perfect recalls of all stored rules is established and two effective encoding algorithms, namely orthogonal encoding and weighted encoding, are devised. The performance of the new model is reported and compared with that of the FAM model through numerous examples.>