On the Role of Similarity in Analogical Transfer

Abstract

Analogical transfer consists in making the assumption that if two situations are alike in some respect, they may be alike in others. This paper studies analogical transfer from the viewpoint of measurement theory. We show that analogical transfer can be seen as a similaritybased reasoning on "contour lines" in a qualitative similarity model: if a contour line is included in another at a given point, then it may also be included at other points that are not too (analogically) dissimilar.

Case-based prediction – A survey

“…A rule (σ S = α) → (σ R = β) is a piece of knowledge that states that σ R is compatible with σ S when σ S takes the value α, and that the resulting similarity level for σ R is β. Potential outcomes r for the new case are derived by triggering such rules on pairs of cases involving the new case in a form of similarity-based inference (SBI), by applying variants of the modus ponens schema [11,16,26,98]: for a retrieved case c s = (s, r s ) and a potential new case ĉt = (t, r), triggering the rule (σ S = α) → (σ R = β) on the pair of cases (c s , ĉt ) is of the form…”

“…Even a very specific rule, when learned with a support of 1 as in the previous paragraph, may be dubious, because it amounts to single instance induction. Several works [11,17,28,91] have emphasized the idea that the rules ( σ S = α) → ( σ R = β) that are reasoned upon should be functional dependencies, i.e., have a confidence value of 1 on any two pairs of cases of the case base. From this observation, [5] defined a variation as any function that associates a value to a pair of situations, and a co-variation as a functional dependency between variations.…”