Complex function of a few features of many inputs

  
Figure 18: Examples of preprocessing high dimensional data for locally weighted learning

Some types of data are very high dimensional. Examples of this are images, acoustic signals, and time series data. In each case, the data is often pre-processed to extract relevant features (as in fig. 18). Common feature extraction routines include principal components analysis, transformation to the frequency domain, and histogramming. The results of the feature extraction often yield a data set like those shown in section 4.3. Examples of this kind of application are:

• Mapping from acoustic signals to ``probability of machine breakdown''
• Time series data analysis
• Mapping from images to classifications
• Product inspection, medical imagery, thin film imaging