Meaning-based handling of don't care attributes in artificial neural networks

An approach is presented for handling don't care attribute values in artificial neural networks (ANNs), based on interpreting the meaning of the don't care values in the context of the general meaning of the concepts learned by the network. The application of these ideas in the context of a particular ANN formalism (M-RCE networks) is discussed. Several considerations on using these ideas with other ANN formalisms are briefly presented.<>