I am interested in finding out what types of Artificial Neural Networks people actually are using in games. I'm sure most of it is a variant on linear or backprop nets because they are relatively straightforward. I would like to find out if anyone uses anything else like a CMAC (Cerebeller Model Arithmetic Computer), RBF network (Radial Basis Function), Fuzzy Network, or any other variation. So I guess my question is this, what kind of NN have you used or seen used?

All the ANNs I know about have been either simple perceptrons (Black & White for example) or single hidden layer feed forward networks.