Channel

Consider the sign language domain discussed in Section 2.1.2. There are different sources of information, each of which we sample at each time interval - things like the x position, the thumb bend and so on. Each of these ``sources'' of information we term a channel. In Figure 2.1, each of the sub-graphs represents a channel. The Tech Support domain contains a single channel, that of the volume of the conversation.

Formally, a channel can be defined as a function $c$ which maps from a set of timestamps $T$ to a set of values, $V$, i.e.

$$c: T \rightarrow V$$

$T$ can be thought of as the set of times for which the value of the channel $c$ is defined. For simplicity in this thesis, we assume that:

$$T = [0, 1, .., t_{\mathit{max}}]$$

In the Tech Support domains, $t_{max}$ varies between 9 and 14. Looking at Figure 2.1 from the sign language domain, we can see that $t_{max} = 40$.

The set of values $V$ obviously depends on the classification task. In the Tech Support domain, $V=\{L,H\}$. In the sign language domain, $V$ for the fingers is $[0,1]$, but for the x, y and z position may be $[-1.5,1.5]$, since the x, y and z positions (measured in metres) can take these values.