Automatic Design of Efficient Algorithms for Black-Box Optimization in Arbitrary Search Domains

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K16063
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Soft computing
• Research Institution: Shinshu University
• Principal Investigator: AKIMOTO Youhei 信州大学, 学術研究院工学系, 助教 (20709654)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 進化計算 / ハイパーパラメータ / 高次元最適化 / 情報幾何 / 制約付き最適化 / 収束率解析 / 最適パラメータ同定

Outline of Final Research Achievements

In this research, we developed and examined a methodology of optimization methods that drastically improves optimization process for black box optimization, which is indispensable for solving engineering, energy and environmental problems. Considering that the bottleneck of optimization process is often hyper-parameter tuning based on trial-and-error, we tried to remove the need of the hyper-parameter tuning, so that the users can use the search algorithm out-of-the-box. We experimentally and theoretically analyzed the influence on hyper-parameter of search algorithms, and succeeded to derive the optimum values of some parameters. We also developed an adaptation mechanism for other preliminary parameters and succeeded in removing the hyper-parameter adjustment. In addition, we adopted statistical techniques and heuristic techniques to improve the search efficiency of methodology itself. The results of this research are expected to lead to the diffusion of optimization technology.

Free Research Field

ブラックボックス最適化