This paper presents Natural Evolution Strategies (NES), a recent family of
algorithms that constitute a more principled approach to black-box optimization
than established evolutionary algorithms. NES maintains a parameterized
distribution on the set of solution candidates, and the natural gradient is
used to update the distribution's parameters in the direction of higher
expected fitness. We introduce a collection of techniques that address issues
of convergence, robustness, sample complexity, computational complexity and
sensitivity to hyperparameters. This paper explores a number of implementations
of the NES family, ranging from general-purpose multi-variate normal
distributions to heavy-tailed and separable distributions tailored towards
global optimization and search in high dimensional spaces, respectively.
Experimental results show best published performance on various standard
benchmarks, as well as competitive performance on others.
%0 Journal Article
%1 wierstra2011natural
%A Wierstra, Daan
%A Schaul, Tom
%A Glasmachers, Tobias
%A Sun, Yi
%A Schmidhuber, Jürgen
%D 2011
%K evolution-strategies genetic-programming
%T Natural Evolution Strategies
%U http://arxiv.org/abs/1106.4487
%X This paper presents Natural Evolution Strategies (NES), a recent family of
algorithms that constitute a more principled approach to black-box optimization
than established evolutionary algorithms. NES maintains a parameterized
distribution on the set of solution candidates, and the natural gradient is
used to update the distribution's parameters in the direction of higher
expected fitness. We introduce a collection of techniques that address issues
of convergence, robustness, sample complexity, computational complexity and
sensitivity to hyperparameters. This paper explores a number of implementations
of the NES family, ranging from general-purpose multi-variate normal
distributions to heavy-tailed and separable distributions tailored towards
global optimization and search in high dimensional spaces, respectively.
Experimental results show best published performance on various standard
benchmarks, as well as competitive performance on others.
@article{wierstra2011natural,
abstract = {This paper presents Natural Evolution Strategies (NES), a recent family of
algorithms that constitute a more principled approach to black-box optimization
than established evolutionary algorithms. NES maintains a parameterized
distribution on the set of solution candidates, and the natural gradient is
used to update the distribution's parameters in the direction of higher
expected fitness. We introduce a collection of techniques that address issues
of convergence, robustness, sample complexity, computational complexity and
sensitivity to hyperparameters. This paper explores a number of implementations
of the NES family, ranging from general-purpose multi-variate normal
distributions to heavy-tailed and separable distributions tailored towards
global optimization and search in high dimensional spaces, respectively.
Experimental results show best published performance on various standard
benchmarks, as well as competitive performance on others.},
added-at = {2019-09-19T16:20:53.000+0200},
author = {Wierstra, Daan and Schaul, Tom and Glasmachers, Tobias and Sun, Yi and Schmidhuber, Jürgen},
biburl = {https://www.bibsonomy.org/bibtex/2922659dfceb5b1d7ada066a65fe31d76/kirk86},
description = {[1106.4487] Natural Evolution Strategies},
interhash = {52c996f36a2a05bbff84430a9b729939},
intrahash = {922659dfceb5b1d7ada066a65fe31d76},
keywords = {evolution-strategies genetic-programming},
note = {cite arxiv:1106.4487},
timestamp = {2019-09-19T16:20:53.000+0200},
title = {Natural Evolution Strategies},
url = {http://arxiv.org/abs/1106.4487},
year = 2011
}