In machine learning, a margin often refers to the distance between the two hyperplanes that separate linearly-separable classes of data points.
The term is most commonly used when discussing support vector machines, but often appears in other literature discussing boundaries between points in a vector space.