In recent years modular neural networks have became more popular for applications in various areas, some examples are presented here.
In the article [kwon94] the usage of a modular neural network for function approximation is described.
A stock market prediction system using MNN is presented in [kimo90]. The system is based on expert modules, realized as MLFFNs. Each expert has its own input domain and preprocessing unit. A final postprocessing unit combines the results from all the modules to an overall output.
An adaptive MNN for character recognition is introduced in [mui94], and it is demonstrated that the modular topology has a high fault tolerance.
Bellotti et. al. describe a network consisting of a self organizing map and a MLP [bell95]. It is used to separate the signal from the background noise in a cosmic ray space experiment.
A similar modular structure is suggested in [blon93]. The task is the classification of `remote senced data'. The architecture uses an unsupervised module to compress the high dimensional input data and a MLP as a classifier.
A description of a patient independent ECG recognition system based on a modular connectionist NN is found in [farr93]. A combination of associative and competitive learning is used.