Case Based Systems

• They say that RBSs are good when there is a theory for the domain. That is, you have rules.
• When you don't have theory, you can use cases.
• Roughly, you describe the problem with a series of cases with answers.
• You get the cases from experts who've solved problems in the domain, and they should be real cases.
• You encode the cases as a set of feature-value pairs. (E.g. a garden.) (See frames.)
• You ask a question by presenting a case.
• The system then finds the nearest case, and gives you that answer.
• Nearness is a difficult thing, but you can use distance if the values are ordinal.
• This also fits naturally into state spaces. The number of possible cases is the size of the space.
• Nearness means that using your nearness metric, similar cases really do have the same solution.