Log-Linear Convergence and Optimal Bounds for the (1+1)-ES
نویسندگان
چکیده
The (1 + 1)-ES is modeled by a general stochastic process whose asymptotic behavior is investigated. Under general assumptions, it is shown that the convergence of the related algorithm is sub-log-linear, bounded below by an explicit log-linear rate. For the specific case of spherical functions and scale-invariant algorithm, it is proved using the Law of Large Numbers for orthogonal variables, that the linear convergence holds almost surely and that the best convergence rate is reached. Experimental simulations illustrate the theoretical results.
منابع مشابه
Log-linear Convergence of the Scale-invariant (μ/μw, λ)-ES and Optimal μ for Intermediate Recombination for Large Population Sizes
Evolution Strategies (ESs) are population-based methods well suited for parallelization. In this report, we study the convergence of the (μ/μw, λ)-ES, an ES with weighted recombination, and derive its optimal convergence rate and optimal μ especially for large population sizes. First, we theoretically prove the log-linear convergence of the algorithm using a scale-invariant adaptation rule for ...
متن کاملLog-Linear Convergence of the Scale-Invariant (µ/µw, lambda)-ES and Optimal µ for Intermediate Recombination for Large Population Sizes
Evolution Strategies (ESs) are population-based methods well suited for parallelization. In this paper, we study the convergence of the (μ/μw, λ)-ES, an ES with weighted recombination, and derive its optimal convergence rate and optimal μ especially for large population sizes. First, we theoretically prove the log-linear convergence of the algorithm using a scale-invariant adaptation rule for t...
متن کاملConvergence results for the (1, )-SA-ES using the theory of -irreducible Markov chains
This paper investigates theoretically the (1, )-SA-ES on the well known sphere function. We prove sufficient conditions on the parameters of the algorithm ensuring the convergence of 1/n ln(||Xn||), where Xn is the parent at generation n. This in turn guarantees the asymptotic log-linear convergence or divergence of the algorithm. The technique used for this analysis calls upon the theory of Ma...
متن کاملConvergence Rates For The Distribution Of Program Outputs
Fitness distributions (landscapes) of programs tend to a limit as they get bigger. Markov chain convergence theorems give general upper bounds on the linear program sizes needed for convergence. Tight bounds (exponential in N , N logN and smaller) are given for five computer models (any, average, cyclic, bit flip and Boolean). Mutation randomizes a genetic algorithm population in 1 4 (l + 1)(lo...
متن کاملcient Parallel Factorization and Solution ofStructured and Unstructured Linear
This paper gives improved parallel methods for several exact factoriza-tions of some classes of symmetric positive deenite (SPD) matrices. Our factorizations also provide us similarly eecient algorithms for exact computation of the solution of the corresponding linear systems (which need not be SPD), and for nding rank and determinant magnitude. We assume the input matrices have entries that ar...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2007