A pattern space is a graphical representation of a set of input patterns and their desired responses (Dawson, 2004). Each input pattern is represented as a point in this space. The coordinates of each pattern are obtained by treating each input unit’s activity as a coordinate value. As a result, the number of input units determines the dimensionality of the pattern space. One can plot the patterns in this space, and then use symbols or colors to represent different types of patterns. This is important because then one can examine how the space needs to be divided up or partitioned (e.g., by output or by hidden units) to separate all of the patterns of one type form all of the patterns of another. This can be used to assess what type of network architecture is required to solve a pattern classification problem, or to prove that some problem is beyond the ken of a particular type of network.