Generality

The architecture appears to be general enough to apply in many domains. We can provide domain-specific knowledge through:

• Selecting appropriate PEPs (parametrised event primitives) for the domain. For example, in some domains we may know that a particular type of event is likely to occur - for example, humps, peaks, changes in level, sudden bursts of noise. We can take advantage of this. Furthermore, we are not constricted to one single PEP, we can use several different ones and choose the ones that perform the best. If we do not have domain knowledge, we have several ``fallback'' PEPs that may be useful, such as piecewise polynomial models, maxima and minima etc. Preliminary results indicate that these simple PEPs work well as defaults.
• Selecting the clustering algorithm used. We may use domain knowledge about the parameter space that is likely to occur to adjust the parameters of an algorithm, or select an appropriate algorithm.
• Selecting the appropriate feature selection and classification algorithms.