Information-Theoretic Connectionist Networks
The process of knowledge discovery can be automated by using the information-theoretic fuzzy approach, termed IFN for Information Fuzzy Network. According to this approach, the interactions between input and target attributes of any type (discrete, continuous, etc.) are represented by an information-theoretic connectionist network. Each internal node of an information-theoretic network stands for a single attribute-value or conjunction of attribute-values. The weights of inter-node connections express the amount and form of association between the internal nodes and the values of the target attribute. This is a local data model, since each node / conjunction of attribute-values represents a subset of the input space. As explained in Chapter 1 above, a local model is supposed to provide a more accurate (but not necessarily a simpler) description of data than a global model. Some ways of simplifying and generalizing an information-theoretic model will be discussed later in Chapter 4. The network fuzzification aimed at detecting unreliable data is covered in Chapter 8 (section 8.2).
KeywordsHide Layer Mutual Information Hide Node Connection Weight Target Node
Unable to display preview. Download preview PDF.