A stochastic conjugate gradient method for the approximation of functions (Q765306)

Conjugate gradient methods are most useful for computations such as approximations and optimization algorithms. In this article, a stochastic approach is being used in connection with the conjugate gradient method by replacing inner products by stochastic sampling and avoiding the storage of the covariance matrices of the normal equations for the least squares problem. The method is described together with a convergence analysis and a discussion of the implementation.