A Study of Parallel Evolutionary Algorithm Independent to Evaluation Time Variances

2020 Fiscal Year Final Research Report

• Project/Area Number: 19K20362
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61040:Soft computing-related
• Research Institution: Tokyo Metropolitan University
• Principal Investigator: Harada Tomohiro 東京都立大学, システムデザイン研究科, 助教 (40755518)
• Project Period (FY): 2019-04-01 – 2021-03-31
• Keywords: 進化計算 / 並列計算 / 最適化

Outline of Final Research Achievements

This research proposes an efficient parallelization method for evolutionary algorithms (PEA), which are typical methods for solving optimization problems. Conventional PEA has a problem of not obtaining the optimal solution in a short computation time when the evaluation time of solutions differs and is biased. To address this problem, this research has proposed the following three PEA approaches; (1) semi-asynchronous PEA that can arbitrarily set the asynchrony of parallelization during optimization, (2) asynchronous PEA that introduces a selection mechanism considering the search progress of solutions, and (3) synchronous PEA that improves the computer utilization rate by the precedence evaluation.

Free Research Field

知能情報学

Academic Significance and Societal Importance of the Research Achievements

本研究は，代表的な最適化手法である進化的アルゴリズム（EA）の計算時間削減のための効果的な並列化手法を確立した．実世界の多くの最適化問題では，解候補の評価にシミュレーションや複雑な数値計算が必要になるなど，莫大な計算時間が必要になり，かつそれぞれの解候補の評価時間は不均一である．これに対し，本研究の研究成果によって，EAの最適化性能を低下させることなく，最適化問題に要する計算時間を大幅に削減でき，製品開発やサービス提供のプロセスを大きく加速できる．