Learning and mathematical reasoning using adaptive signal processing techniques

The MLS (machine learning system) approach offers a more generic solution to the problem of fault diagnosis and adjustments in electronic devices and systems than the classical ESs. The MLS is capable of learning the relationships between the inputs and the outputs of a system by looking at a number of examples which include features and corresponding desired action. Therefore the performance of the MLS depends on how the features have been selected. The authors present a technique for finding the correlation between the selected features, the significance of each individual feature and the interaction between the outcomes. This enables one to modify the feature set and also the training sequence. The main limitations of this approach are (a) the combiners are a linear structure, therefore they can not be used to model nonlinear relationships and (b) the decisions of the DAP are based on the estimates of the correlation matrices. In order to resolve the first problem it is intended to look at nonlinear structures such as neural networks for the implementation of the MLS. The second problem may be solved by either using a large number of training examples or introducing some form of memory into the structure of the DAP so that the present decisions are not only based on the present correlation matrices but also on the previous decisions