Stochastic minimization with adaptive memory (Q1893594)

A variant of a stochastic minimization algorithm is studied. A common feature of random search algorithms is that little or no use is made of information on the local structure of the function to be minimized. A random search algorithm with adaptive memory is presented which is characterized by the use of an adaptive Gaussian memory for biasing the exploration. It is shown how the random search can be supplemented with a simple, dynamic memory mechanism, enabling the algorithm to maintain a more global view of the function to be minimized and to use it as a predictive tool in the search for the minimum. Several problems are considered to demonstrate the increased efficiency of the algorithm over the basic variant. Comparison with classic minimization techniques, such as the conjugate gradient and Levenberg- Marquardt methods, are also given. Of particular interest is the application of the presented algorithm to the computation of the minimum eigenvalue of a singular differential operator on a Hilbert space for which traditional techniques perform badly.