Self-Adjusting Population Sizes for Non-Elitist Evolutionary Algorithms: Why Success Rates Matter

被引:19
作者
Fajardo, Mario Alejandro Hevia [1 ]
Sudholt, Dirk [2 ]
机构
[1] Univ Sheffield, Dept Comp Sci, Sheffield, S Yorkshire, England
[2] Univ Passau, Chair Algorithms Intelligent Syst, Passau, Germany
来源
PROCEEDINGS OF THE 2021 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE (GECCO'21) | 2021年
关键词
Parameter control; Theory; Runtime analysis; Non-elitism; Drift analysis; Evolutionary algorithms; BLACK-BOX COMPLEXITY;
D O I
10.1145/3449639.3459338
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent theoretical studies have shown that self-adjusting mechanisms can provably outperform the best static parameters in evolutionary algorithms on discrete problems. However, the majority of these studies concerned elitist algorithms and we do not have a clear answer on whether the same mechanisms can be applied for non-elitist algorithms. We study one of the best-known parameter control mechanisms, the one-fifth success rule, to control the offspring population size A in the non-elitist (1, lambda) EA. It is known that the (1, lambda) EA has a sharp threshold with respect to the choice of A where the runtime on ONEMAx changes from polynomial to exponential time. Hence, it is not clear whether parameter control mechanisms are able to find and maintain suitable values of lambda. We show that the answer crucially depends on the success rate s (i. e. a one-(s + 1)-th success rule). We prove that, if the success rate is appropriately small, the self-adjusting (1, lambda) EA optimises ONEMAX in O(n) expected generations and O(n log n) expected evaluations. A small success rate is crucial: we also show that if the success rate is too large, the algorithm has an exponential runtime on ONEMAX.
引用
收藏
页码:1151 / 1159
页数:9
相关论文
共 33 条
  • [1] Badkobeh G, 2014, LECT NOTES COMPUT SC, V8672, P892
  • [2] BenjaminDoerr CarolaDoerr., 2019, P GEN EV COMP GECCO
  • [3] Böttcher S, 2010, LECT NOTES COMPUT SC, V6238, P1, DOI 10.1007/978-3-642-15844-5_1
  • [4] Self-Adaptation in Nonelitist Evolutionary Algorithms on Discrete Problems With Unknown Structure
    Case, Brendan
    Lehre, Per Kristian
    [J]. IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION, 2020, 24 (04) : 650 - 663
  • [5] Doerr B., 2020, Theory of Evolutionary Computation: Recent Developments in Discrete Optimization, P271, DOI [DOI 10.1007/978-3-030-29414-4, 10.1007/978-3-030-29414-4]
  • [6] Optimal parameter choices via precise black-box analysis
    Doerr, Benjamin
    Doerr, Carola
    Yang, Jing
    [J]. THEORETICAL COMPUTER SCIENCE, 2020, 801 (801) : 1 - 34
  • [7] On the Runtime Analysis of Selection Hyper-Heuristics with Adaptive Learning Periods
    Doerr, Benjamin
    Lissovoi, Andrei
    Oliveto, Pietro S.
    Warwicker, John Alasdair
    [J]. GECCO'18: PROCEEDINGS OF THE 2018 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE, 2018, : 1015 - 1022
  • [8] Doerr B, 2019, ALGORITHMICA, V81, P593, DOI 10.1007/s00453-018-0502-x
  • [9] Doerr B, 2018, ALGORITHMICA, V80, P1658, DOI 10.1007/s00453-017-0354-9
  • [10] Provably Optimal Self-adjusting Step Sizes for Multi-valued Decision Variables
    Doerr, Benjamin
    Doerr, Carola
    Koetzing, Timo
    [J]. PARALLEL PROBLEM SOLVING FROM NATURE - PPSN XIV, 2016, 9921 : 782 - 791