| Global optimizers are powerful tools for searching complicated function spaces such as those found in modern high-fidelity engineering models. These models provide increasingly detailed insights into system behaviors, but are often expensive to evaluate and difficult to search. While techniques exist for solving global optimization problems there is still room for developing faster, more reliable, and easier to implement algorithms. This thesis investigates the design of a global optimization algorithm, with a focus on the search strategies of gradient descent, memory, and multiresolution search. Major contributions presented in this thesis include the proposal of a regional gradient, the examination of the use of regional gradients in global search, and the proposal and development of a novel optimization algorithm. Additionally, a case study using the multiresolution estimated gradient architecture (MEGA) algorithm to solve a classification problem is presented. |
