A local explanation is a form of explainable AI, which attempts to explain the behaviour of a model on a specific input–output. For example LIME builds a linear discriminant based on the particular input–output and other examples near to it. This matches the behavour of the model in the locality of the partcular input–output, but is not expected to hold for inputs that are at all different. This is in contrast to a global explanation, which attempts to explain the entire space of input–output behaviour of the model.
Used in glossary entries: explainable AI, global explanation, LIME, linear discriminant analysis