Adaptive global optimization algorithms

Global optimization is concerned with finding the minimum value of a function where many local minima may exist. The development of a global optimization algorithm may involve using information about the target function (e.g., differentiability) and functions based on statistical models to better the worst case time complexity and expected error of similar deterministic algorithms. Recent algorithms are investigated, new ones proposed and their performance is analyzed. Minimum, maximum and average case error bounds for the algorithms presented are derived. Software architecture implemented with MATLAB and Java is presented and experimental results for the algorithms are displayed. The graphical capabilities and function-rich MATLAB environment are combined with the object oriented features of Java, hosted on the computer system described in this paper, to provide a fast, powerful test environment to provide experimental results. In order to do this, matlabcontrol, a third party set of procedures that allows a Java program to call MATLAB functions to access a function such as voronoi() or to provide graphical results, is used. Additionally, the Java implementation can be called from, and return values to, the MATLAB environment. The data can then be used as input to MATLAB\u27s graphing or other functions. The software test environment provides algorithm performance information such as whether more iterations or replications of a proposed algorithm would be expected to provide a better result for an algorithm. It is anticipated that the functionality provided by the framework would be used for initial development and analysis and subsequently removed and replaced with optimized (in the computer efficiency sense) functions for deployment