NEW DECISION RULES IN STATISTICAL PATTERN RECOGNITION

In this paper, one combines information theory, and more especially the concept of entropy, with the statistical theory of decision to derive new criteria for pattern recognition. A generalized definition of entropy is considered as a risk function, and the generalized decision rules so obtained contain the family of the Bayesian decisions as special cases. These criteria may help to check the results obtained by usual techniques; they can be used in adaptive and learning systems, and more generally they can be useful in cybernetic systems.