Representing time explicitly

An examination of domains where temporal classification arises indicates that signals that are of the same class tend to vary in the following ways:

• There are random amplitude variations in the channels - what would normally be classified as noise. Also, it may be that a similar-shaped event occurs, but that it has a greater or lesser amplitude as a whole.
• The starting point (often called the ``onset'') of an event may be different, even though two events may look very similar. An example of this was shown in figure 3.1. Stream I and II would be considered by most to be members of the same class of shapes, even though the ``hump'' occurs at a different time.
• The duration of an event may be longer or shorter, even though the event may otherwise be the same.

If we can represent the data in a way that explicitly models these events and the variations in their amplitudes, their onset and their durations, then we may be able to convert the problem into one more palatable to existing classification techniques. This would simultaneously reduce the data we had to deal with, while also making comparisons between streams in a way that is intuitively more appealing.