Implementation of a neural net expert system in the presence of don´t care and don´t know features

The implementation of a neural network acting as an expert system is discussed. In some cases a particular feature may not be appropriate to describe a particular object (don´t care condition) or a particular feature cannot be extracted or is missing (don´t know feature). It is shown how the traditional feedforward neural network cannot be trained on these types of features, and an alternate structure is shown that is optimal when the features are discrete and independent and errors are independent. The structure is highly regular. It is shown how to implement the network using operational amplifiers and logic gates. The application to a real-world problem of identifying ordnances is shown