Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Drift Analysis with Fitness Levels for Elitist Evolutionary Algorithms

Jun He, Yuren Zhou

TL;DR

This work addresses the longstanding question of how to obtain tight hitting-time bounds for elitist evolutionary algorithms using fitness-level transitions. It combines drift analysis with fitness-level partitions to derive the tightest metric bounds and then converts these into general linear bounds with a flexible family of coefficients $c_{k,\ell}$. The authors prove that the tightest metric bounds are achieved when the drift inequalities become equalities and show how to construct linear bounds that subsume existing Type-$0$, Type-$1$, $c$, and $c_\ell$ bounds; they further demonstrate the framework on the (1+1) EA for OneMax and TwoMax1, revealing that Type-$c_{k,\ell}$ bounds can be tight on landscapes with shortcuts. The framework provides a generic, scalable approach to obtain tight worst-case runtime bounds for elitist EAs across landscapes, including those with shortcuts, and suggests practical strategies for coefficient computation.Overall, the paper advances a unified, adaptable methodology for precise time-bound analysis in evolutionary computation.

Abstract

The fitness level method is a popular tool for analyzing the hitting time of elitist evolutionary algorithms. Its idea is to divide the search space into multiple fitness levels and estimate lower and upper bounds on the hitting time using transition probabilities between fitness levels. However, the lower bound generated by this method is often loose. An open question regarding the fitness level method is what are the tightest lower and upper time bounds that can be constructed based on transition probabilities between fitness levels. To answer this question, {\color{red} we combine drift analysis with fitness levels and define the tightest bound problem as a constrained multi-objective optimization problem subject to fitness levels.} The tightest metric bounds from fitness levels are constructed and proven for the first time. Then linear bounds are derived from metric bounds and a framework is established that can be used to develop different fitness level methods for different types of linear bounds. The framework is generic and promising, as it can be used to draw tight time bounds on both fitness landscapes without and with shortcuts. This is demonstrated in the example of the (1+1) EA maximizing the TwoMax1 function

Drift Analysis with Fitness Levels for Elitist Evolutionary Algorithms

TL;DR

This work addresses the longstanding question of how to obtain tight hitting-time bounds for elitist evolutionary algorithms using fitness-level transitions. It combines drift analysis with fitness-level partitions to derive the tightest metric bounds and then converts these into general linear bounds with a flexible family of coefficients . The authors prove that the tightest metric bounds are achieved when the drift inequalities become equalities and show how to construct linear bounds that subsume existing Type-, Type-, , and bounds; they further demonstrate the framework on the (1+1) EA for OneMax and TwoMax1, revealing that Type- bounds can be tight on landscapes with shortcuts. The framework provides a generic, scalable approach to obtain tight worst-case runtime bounds for elitist EAs across landscapes, including those with shortcuts, and suggests practical strategies for coefficient computation.Overall, the paper advances a unified, adaptable methodology for precise time-bound analysis in evolutionary computation.

Abstract

The fitness level method is a popular tool for analyzing the hitting time of elitist evolutionary algorithms. Its idea is to divide the search space into multiple fitness levels and estimate lower and upper bounds on the hitting time using transition probabilities between fitness levels. However, the lower bound generated by this method is often loose. An open question regarding the fitness level method is what are the tightest lower and upper time bounds that can be constructed based on transition probabilities between fitness levels. To answer this question, {\color{red} we combine drift analysis with fitness levels and define the tightest bound problem as a constrained multi-objective optimization problem subject to fitness levels.} The tightest metric bounds from fitness levels are constructed and proven for the first time. Then linear bounds are derived from metric bounds and a framework is established that can be used to develop different fitness level methods for different types of linear bounds. The framework is generic and promising, as it can be used to draw tight time bounds on both fitness landscapes without and with shortcuts. This is demonstrated in the example of the (1+1) EA maximizing the TwoMax1 function
Paper Structure (24 sections, 24 theorems, 98 equations, 1 figure, 3 tables, 1 algorithm)

This paper contains 24 sections, 24 theorems, 98 equations, 1 figure, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. New research and main results in this paper
  4. Preliminary
  5. Elitist EAs and Markov chains
  6. Hitting time and drift analysis
  7. Fitness level partition and transition probabilities between fitness levels
  8. Shortcuts
  9. Review and discussion of existing fitness level methods
  10. Existing fitness level methods
  11. Case Study 1: A loose Type-$c$ lower bound for the (1+1) EA on TwoMax1
  12. Case Study 2: A loose Type-$c_\ell$ lower bound for the (1+1) EA on TwoMax1
  13. Metric bounds from fitness levels
  14. Drift analysis with fitness levels
  15. Metric bounds
  16. ...and 9 more sections

Key Result

Lemma 1

he2003towards If for any $X\notin S_{\mathrm{opt}}$, the drift $\Delta d(X) \le 1$, then the mean hitting time $m(X) \ge d(X)$.

Figures (1)

  • Figure 1: Left: the (1+1) EA on OneMax$(x)$ where $n=10$. Right: The (1+1) EA on TwoMax1$(x)$ where $n=10$. Dotted lines represent transitions. Solid lines are two shortcuts: $S_6 \to S_0$ skipping $S_1, \cdots, S_5$, and $S_9\to S_5$ skipping $S_6$.

Theorems & Definitions (35)

  • Definition 1
  • Definition 2
  • Definition 3
  • Lemma 1
  • Lemma 2
  • Definition 4
  • Definition 5
  • Proposition 1
  • Proposition 2
  • Proposition 3
  • ...and 25 more