A new probabilistic relaxation method based on probability space partition

摘要

Probabilistic relaxation is an iterative and parallel process, which can be regarded as a recurrent dynamical system. Since the stable states of the system depend on system parameters, one can design a system by estimating or selecting a set of parameters based on incomplete information of a given problem to implement various information processing procedures, such as classification and recognition. In this paper, a new probabilistic relaxation method is proposed based on the elementary theories of both conditional probability and probability space partition. The dynamics of the system and the relationship between convergence properties and system parameters are considered analytically and numerically. The comparison of convergence properties between the proposed method and some existing schemes are provided. The results show that our scheme is more clear and can be easily generalized. © 1997 Pattern Recognition Society.