Impact Statement:Real-world optimization problems, such as aerodynamic design of turbine engines and automated trading, have been successfully solved by metaheuristics. However, practitio...Show More
Abstract:
As revealed by the no free lunch theorem, no single algorithm can outperform any others on all classes of optimization problems. To tackle this issue, methods for recomme...Show MoreMetadata
Impact Statement:
Real-world optimization problems, such as aerodynamic design of turbine engines and automated trading, have been successfully solved by metaheuristics. However, practitioners are confronted with the challenge of how to choose an appropriate metaheuristic algorithm to solve a particular instance of these problems. This paper proposes a recommender system that can automatically select a best-suited metaheuristic algorithm without trial and error on a given problem. The proposed method develops a generic tree-like data structure for representing the difficulties of optimization problems and then trains a deep recurrent neural network to learn to choose the best metaheuristic algorithm, making automated algorithm recommendation practical for real-world problem-solving. The method will make metaheuristic optimization techniques accessible to industrial practitioners, policy makers, and other stakeholders who have no knowledge in metaheuristic algorithms.
Abstract:
As revealed by the no free lunch theorem, no single algorithm can outperform any others on all classes of optimization problems. To tackle this issue, methods for recommending an existing algorithm for solving given problems have been proposed. However, existing recommendation methods for continuous optimization suffer from low practicability and transferability, mainly due to the difficulty in extracting features that can effectively describe the problem structure and lack of data for training a recommendation model. This work proposes a generic recommender system to address the above two challenges. First, a novel method is proposed to represent an analytic objective function of a continuous optimization problem as a tree, which is directly used as the features of the problem. For black-box optimization problems whose objective function is unknown, a symbolic regressor is adopted to estimate the tree structure. Second, a large number of benchmark problems are randomly created based o...
Published in: IEEE Transactions on Artificial Intelligence ( Volume: 1, Issue: 1, August 2020)