Prior symmetry, similarity-based reasoning, and endogenous categorization (Q629323)

The reviewed paper deals with a rational decision-making method consisting in the similarity-based clustering of the objects of interest. The author considers the problem of sequential learning regarding objects of incompletely known properties. Their unknown properties are estimated by means of analogy between the elements of particular clusters (called categories). Attention is mainly paid to the investigation of the considered environment whose elements are to be categorized, and the main results demonstrate the high importance of the concept of symmetry defined in the considered environment. It is shown the close relation between this symmetry and optimality of the categorization-based decision-making procedure of an inference problem. Optimality of the suggested procedure is illustrated by experimental observations.