The logical approach to neural computing is different from the connectionist ideas for NN. The accumulated knowledge is stored as parts of patterns. This neural system has no weights.
The learning algorithm is very simple, the patterns are presented to the inputs of the network and then the patterns are stored in a certain way. This feature is very desirable from the application point of view, because it is fast to train and easy to use. However it lacks biological and psychological plausibility.
From the point of view of a biologist this kind of neural network has very few communalities with the biological description of the nervous system.
However, the architecture viewed from a more abstract point has some points in common with the biological model. The task is performed by many simple units; their only function is to store one piece of information for one input pattern. The ability to generalize is built in to the architecture of the network.
A major advantage of this approach is the ease of implementing the system in conventional hardware. One example of a hardware implementation is the WISARD, an adaptive pattern recognition device [alek95, p73ff,].