Metaheuristic exploration via deep learning object localization

Zang, Mengyu

Honours

2021

Bolufe-Rohler, Antonio

Bachelor of Science

Faculty of Science. Honours in Computer Science

Computer Science

Mathematical and Computational Sciences

University of Prince Edward Island

Charlottetown, PE

Heuristic and metaheuristic optimization algorithms have enjoyed success as the method of choice for solving many real world problems due to their flexibility and speed. Optimization requires that a metaheuristic perform both exploration and exploitation. The role of exploration is to find the most promising region of a search space; recent publications have shown that exploration is the most critical and challenging part of the optimization process. In this research we present an entirely new Show moreHeuristic and metaheuristic optimization algorithms have enjoyed success as the method of choice for solving many real world problems due to their flexibility and speed. Optimization requires that a metaheuristic perform both exploration and exploitation. The role of exploration is to find the most promising region of a search space; recent publications have shown that exploration is the most critical and challenging part of the optimization process. In this research we present an entirely new approach to exploration based on the use of deep learning. We have successfully represented an objective function as a 2-dimensional image; created two different datasets using the point where the global optimum is located as the target attribute to be predicted. We designed, trained and tested different deep neural network convolutional models for predicting the region where the optimum is located, and we have performed extensive experimentation to illustrate the effectiveness of this approach. Show less

Contact Author

PUBLISHED