Multi-module neural network model for higher order association

A multi-module neural network model for high order association have been proposed. It contains plural functional modules each of which is mutually connected to the neural networks with hidden units in order to improve the performance of recall. The model comprises two different type of networks; fundamental modular network (FMN) and intermediate network (IN). Each FMN is mutually connected to each other by INs and works dynamically in cooperation with other functional modules. In this paper, it is also shown that this model has great ability of recollection, same as fully, mutually connected neural networks. The higher order association between four 2D character dot patterns, a corresponding three/four-character-word pattern and an image indicated by the word mean are well demonstrated by the model.