Abstract:For genetic algorithms (GAs) using a bit-string representation of length n, the general recommendation is to take 1/n as mutation rate. In this work, we discuss whether this is justified for multi-modal functions. Taking jump functions and the (1+1) evolutionary algorithm (EA) as the simplest example, we observe that larger mutation rates give significantly better runtimes. For the Jumpm, n function, any mutation rate between 2/n and m/n leads to a speedup at least exponential in m compared to the standard choice. The asymptotically best runtime, obtained from using the mutation rate m/n and leading to a speed-up super-exponential in m, is very sensitive to small changes of the mutation rate. Any deviation by a small (1 ε) factor leads to a slow-down exponential in m. Consequently, any fixed mutation rate gives strongly sub-optimal results for most jump functions. Building on this observation, we propose to use a random mutation rate /n, where is chosen from a power-law distribution. We prove that the (1 + 1) EA with this heavy-tailed mutation rate optimizes any Jumpm, n function in a time that is only a small polynomial (in m) factor above the one stemming from the optimal rate for this m. Our heavy-tailed mutation operator yields similar speed-ups (over the best known performance guarantees) for the vertex cover problem in bipartite graphs and the matching problem in general graphs. Following the example of fast simulated annealing, fast evolution strategies, and fast evolutionary programming, we propose to call genetic algorithms using a heavy-tailed mutation operator fast genetic algorithms.

Fitness Probability Distribution of Bit-Flip Mutation

Mathematical Analysis of Mutation Operator and Its Improved Strategy in Genetic Algorithms

Tight Runtime Bounds for Static Unary Unbiased Evolutionary Algorithms on Linear Functions

Optimal Discrete Uniform Generation from Coin Flips, and Applications

Optimizing Genetic Algorithms Using the Binomial Distribution

A Flexible Evolutionary Algorithm With Dynamic Mutation Rate Archive

An upper bound of the mutation probability in the genetic algorithm for general 0-1 knapsack problem

Dynamic fitness landscapes: expansions for small mutation rates

Fluctuations of fitness distributions and the rate of Muller's ratchet

Fluctuation Theorem for the Mutation Process in in Vitro Evolution

Runtime Analysis for Permutation-based Evolutionary Algorithms

Failed Disruption Propagation in Integer Genetic Programming

On the behavior of the leading eigenvalue of Eigen's evolutionary matrices

Fast genetic algorithms

Evolutionary accessibility of random and structured fitness landscapes

Binary-encoding Genetic Algorithms with Modified Mutation Operation

Survival of the flattest in the quasispecies model

Error thresholds in a mutation-selection model with Hopfield-type fitness

Fourier Analysis Meets Runtime Analysis: Precise Runtimes on Plateaus

Polytopes, graphs and fitness landscapes

Probability Evolutionary Algorithm for Functional and Combinatorial Optimization