Elements of Stochastic Programming

Abstract

In the second part of the chapter is discussed the “potential functions method,” a very fruitful area of the stochastic programming. Even though it is called method, this field is actually a mathematical theory whose practical results are a large number of stochastic recurrent procedures for pattern recognition, approximating algorithms for functions in noisy conditions, development of unified mathematical approach for machine learning on the basis of human preference, proof of the perceptron theory, etc. The stochastic algorithms based on the “potential functions method” have stable convergence and flexibility, and these properties permit fruitful application in utility and value function evaluations and polynomial approximations. The last part of the chapter gives an example of pattern recognition of two sets of positive and negative answers as machine learning procedure.