ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2405.01014
29
2

Proven Runtime Guarantees for How the MOEA/D Computes the Pareto Front From the Subproblem Solutions

2 May 2024
Benjamin Doerr
Martin S. Krejca
Noé Weeks
ArXivPDFHTML
Abstract

The decomposition-based multi-objective evolutionary algorithm (MOEA/D) does not directly optimize a given multi-objective function fff, but instead optimizes N+1N + 1N+1 single-objective subproblems of fff in a co-evolutionary manner. It maintains an archive of all non-dominated solutions found and outputs it as approximation to the Pareto front. Once the MOEA/D found all optima of the subproblems (the ggg-optima), it may still miss Pareto optima of fff. The algorithm is then tasked to find the remaining Pareto optima directly by mutating the ggg-optima. In this work, we analyze for the first time how the MOEA/D with only standard mutation operators computes the whole Pareto front of the OneMinMax benchmark when the ggg-optima are a strict subset of the Pareto front. For standard bit mutation, we prove an expected runtime of O(nNlog⁡n+nn/(2N)Nlog⁡n)O(n N \log n + n^{n/(2N)} N \log n)O(nNlogn+nn/(2N)Nlogn) function evaluations. Especially for the second, more interesting phase when the algorithm start with all ggg-optima, we prove an Ω(n(1/2)(n/N+1)N2−n/N)\Omega(n^{(1/2)(n/N + 1)} \sqrt{N} 2^{-n/N})Ω(n(1/2)(n/N+1)N​2−n/N) expected runtime. This runtime is super-polynomial if N=o(n)N = o(n)N=o(n), since this leaves large gaps between the ggg-optima, which require costly mutations to cover. For power-law mutation with exponent β∈(1,2)\beta \in (1, 2)β∈(1,2), we prove an expected runtime of O(nNlog⁡n+nβlog⁡n)O\left(n N \log n + n^{\beta} \log n\right)O(nNlogn+nβlogn) function evaluations. The O(nβlog⁡n)O\left(n^{\beta} \log n\right)O(nβlogn) term stems from the second phase of starting with all ggg-optima, and it is independent of the number of subproblems NNN. This leads to a huge speedup compared to the lower bound for standard bit mutation. In general, our overall bound for power-law suggests that the MOEA/D performs best for N=O(nβ−1)N = O(n^{\beta - 1})N=O(nβ−1), resulting in an O(nβlog⁡n)O(n^\beta \log n)O(nβlogn) bound. In contrast to standard bit mutation, smaller values of NNN are better for power-law mutation, as it is capable of easily creating missing solutions.

View on arXiv
Comments on this paper