arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2204.00531cs.NEmath.PR
收藏

单调函数上 $(1, \lambda)$-EA 的自适应种群规模

Self-adjusting Population Sizes for the $(1, λ)$-EA on Monotone Functions

Marc Kaufmann, Maxime Larcher, Johannes Lengler, Xun Zou

AI总结:

本文分析采用 $(1:s+1)$ 成功规则自适应种群规模的 $(1,\lambda)$-EA,证明其在单调函数上的效率阈值取决于 $s$,并给出代数与评估复杂度及动态环境下的推广结论。

摘要:

我们研究变异率为 $c/n$(其中 $c\le 1$)的 $(1,\lambda)$-EA,其种群规模由 $(1:s+1)$-成功规则自适应控制。最近,Hevia Fajardo 和 Sudholt 证明,当 $c=1$ 时,该设置在 \onemax 上对于 $s<1$ 是高效的,但当 $s\ge 18$ 时则低效。令人惊讶的是,最困难的部分并不接近最优值,而是处于线性距离处。我们证明这种行为并非 \onemax 所特有。如果 $s$ 较小,该算法在所有单调函数上都是高效的;如果 $s$ 较大,则它在所有单调函数上都需要超多项式时间。在前一种情况下,对于 $c<1$,我们证明代数数的上界为 $O(n)$,函数评估次数的上界为 $O(n\log n)$;对于 $c=1$,我们证明代数数为 $O(n\log n)$,评估次数为 $O(n^2\log\log n)$。我们还形式化地证明,如果算法从最优值附近开始,则无论 $s$ 为何,优化始终很快。所有结果在适应度函数每一代都发生变化的动态环境中同样成立。

英文摘要:

We study the $(1,λ)$-EA with mutation rate $c/n$ for $c\le 1$, where the population size is adaptively controlled with the $(1:s+1)$-success rule. Recently, Hevia Fajardo and Sudholt have shown that this setup with $c=1$ is efficient on \onemax for $s<1$, but inefficient if $s \ge 18$. Surprisingly, the hardest part is not close to the optimum, but rather at linear distance. We show that this behavior is not specific to \onemax. If $s$ is small, then the algorithm is efficient on all monotone functions, and if $s$ is large, then it needs superpolynomial time on all monotone functions. In the former case, for $c<1$ we show a $O(n)$ upper bound for the number of generations and $O(n\log n)$ for the number of function evaluations, and for $c=1$ we show $O(n\log n)$ generations and $O(n^2\log\log n)$ evaluations. We also show formally that optimization is always fast, regardless of $s$, if the algorithm starts in proximity of the optimum. All results also hold in a dynamic environment where the fitness function changes in each generation.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑