ABSTRACT
This section concerns neural networks which are hybrid either in terms of structure or in terms of training algorithms. The counterpropagation network is one that incorporates structural characteristics of the Kohonen and Grossberg networks and it is trained by composite supervised-unsupervised methods. The adaptive critic concept concerns neural network implementations of reinforcement learning where teacher information is available, a supervised learning characteristic, but target outputs are not specified, an unsupervised learning characteristic. The counterpropagation network as well as a number of adaptive critic implementations are taken up in this section.
