The CMA Evolution Strategy: A Comparing Review

By Nikolaus Hansen. 2006

2,322 citations recorded by the source corpus.

Subfields: Developmental Biology, Ecology, Evolution, Behavior and Systematics, Genetics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Developmental Biology

In Ecology, Evolution, Behavior and Systematics

In Genetics

Citations in the corpus

On First-Order Meta-Learning Algorithms (2018, methodology): “Initially, we tried optimizing the Reptile hyper-parameters using CMA-ES [6].”
Fast Bayesian Optimization of Machine Learning Hyperparameters on Large Datasets (2016, methodology): “To maximize the acquisition function we used the blackbox optimizer DIRECT (Jones, 2001) and CMAES (Hansen, 2006).”
One Pixel Attack for Fooling Deep Neural Networks (2017, methodology): “As the tradeoff of being a low-cost, easy-implemented L0 attack, we do not expect one pixel attack can achieve significantly better robustness against such detection methods compared to other L0 attacks such as [32].”
Stein Variational Policy Gradient (2017, methodology): “Instead of computing the ﬁnite difference for each individual parameter, rollout efﬁcient methods such as SPSA (Spall, 1998), PEPG (Sehnke et al., 2010) and evolutionary strategy approximations (Hansen, 2006, Mannor et al., 2003) have been proposed.”
Safe Reinforcement Learning (2015, methodology): “In all of the experiments in this dissertation, we solve (approximately) for this global maximum using the black-box optimization algorithm covariance matrix adaptation evolution strategy (CMA-ES) (Hansen, 2006).”
A Surrogate-Assisted Reference Vector Guided Evolutionary Algorithm for Computationally Expensive Many-Objective Optimization (2018, methodology): “For instance, MOEA/D-EGO uses differential evolution [44] and SMS-EGO employs covariance matrix adaptation [19] to optimize the hyperparameters,”
A Survey of Automatic Parameter Tuning Methods for Metaheuristics (2020, methodology): “In parameter control, self-adaptive parameter control, such as step-size control and covariance matrix adaptation in covariance matrix adaptation ES (CMA-ES) [23], is the state-of-the-art method for numerical parameter control.”
Terrain-adaptive bipedal locomotion control (2010, methodology): “We use the CMA evolution strategy as a black-box optimizer [Hansen 2006].”
Surrogating the surrogate: accelerating Gaussian-process-based global optimization with a mixture cross-entropy algorithm (2010, methodology): “This is similar to CMA-ES ( Hansen, 2006), another eﬃcient search heuristics based on the evolutionary update of a single Gaussian.”
Automated Algorithm Configuration and Parameter Tuning (2012, methodology): “These meth- ods range from well-known numerical optimisation procedures, such as the NelderMead Simplex algorithm (Nelder and Mead, 1965; Bűrmen et al, 2006) or the more recent, gradient-free CMA-ES algorithm (Hansen, 2006; Hansen and Ostermeier, 2001; Hansen and Kern, 2004), to approaches based on”
Simulation optimization: a review of algorithms and applications (2014, methodology): “RAS minimizes the KL-divergence with respect to the distribution in the current iteration, called the reference model. Covariance Matrix Adaptation–Evolution Strategy (CMA-ES). In the CMAES algorithm [81], new samples are obtained from a multivariate normal distribution, and inter-variable dependenc”
Quantum annealing for industry applications: introduction and review (2021, neutral): “They used the covariance matrix adaptation evolutionary strategy [75] to heuristically optimize the annealing offset of the qubit chains, i.e., the Δ t of advance or delay for each qubit in physical time.”
Differentiable Expected Hypervolume Improvement for Parallel Multi-Objective Bayesian Optimization (2020, methodology): “If not, gradients are either approximated (e.g. with ﬁnite differences) or gradient-free methods (e.g. DIRECT [36] or CMA-ES [31]) are used.”