The CMA Evolution Strategy: A Tutorial
By Nikolaus Hansen. 2016
1,717 citations recorded by the source corpus.
This tutorial introduces the CMA Evolution Strategy (ES), where CMA stands for Covariance Matrix Adaptation. The CMA-ES is a stochastic, or randomized, method for real-parameter (continuous domain) optimization of non-linear, non-convex functions. We try to motivate and derive the algorithm from intuitive concepts and from requirements of non-linear, non-convex search in continuous domain.
Subfields: Artificial Intelligence, Optimization & Training, Evolutionary Computation
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Sergey Levine
- Yoshua Bengio
- Quoc V. Le
- Pieter Abbeel
- Jianfeng Gao
- Mohit Bansal
- Percy Liang
- Ruslan Salakhutdinov
In Optimization & Training
- Yoshua Bengio
- Frank Hutter
- Quoc V. Le
- J. Urgen Schmidhuber
- Song Han
- David R Ha
- Hugo Larochelle
- Sergey Levine
In Evolutionary Computation
- J. Urgen Schmidhuber
- David R Ha
- Song Han
- Charles Herrmann
- Frank Hutter
- Nuno Lau
- M. McCourt
- A. Larcher
Citations in the corpus
- Recurrent World Models Facilitate Policy Evolution (2018, methodology): “We used Covariance-Matrix Adaptation Evolution Strategy (CMA-ES) [29] to evolve C’s weights.”
- Hyper-Parameter Optimization: A Review of Algorithms and Applications (2020, methodology): “, 2017), and covariance matrix adaption evolutionary strategy (Hansen, 2016).”
- World Models (2018, methodology): “We used Covariance-Matrix Adaptation Evolution Strategy (CMA-ES) (Hansen, 2016) to evolve the weights for our C Model.”
- Leveraging Pre-trained Checkpoints for Sequence Generation Tasks (2019, methodology): “We then performed a hyperparameter search on the evaluation set using CMA-ES (Hansen, 2016) to find an optimal subset of layers to use.”
- Conditioning by adaptive sampling for robust design (2019, methodology): “…a number of model-based optimization schemes, such as Evolutionary Distribution Algorithm (EDA) and Information Geometric Optimization (IGO) approaches (Hansen, 2006; Ollivier et al., 2017), both of which have shown to have good practical performance on a wide range of optimization problems.”
- Tree-structured Parzen estimator: Understanding its algorithm components and their roles for better empirical performance (2023, methodology): “If the search space has only numerical parameters, a promising candidate would be CMA-ES (Hansen, 2016) for large-budget settings and the Nelder-Mead method (Nelder & Mead, 1965) for small-budget settings.”
- An Asymptotically Optimal Multi-Armed Bandit Algorithm and Hyperparameter Optimization (2020, methodology): “Population-based methods (Hansen, 2016; Jaderberg et al., 2017) such as genetic algorithms, evolutionary strategies, and particle swarm optimization use guided search and usually outperform random search.”
- A review on Deep Reinforcement Learning for Fluid Mechanics (2019, methodology): “The follow-up policy-based optimization (PBO) algorithm relies on 252 a variant of the vanilla policy gradient method and delivers several major improvements by adopting 253 key heuristics from the covariance matrix adaptation evolution strategy (CMA-ES [34]), including the 254 use of a valid, full ”
- GCN-RL Circuit Designer: Transferable Transistor Sizing with Graph Neural Networks and Reinforcement Learning (2020, methodology): “The optimization algorithms such as BO [9], MACE [2], ES [8] consider the circuits as a black box and conduct optimization.”
- Learning search spaces for Bayesian optimization: Another view of hyperparameter transfer learning (2019, methodology): “Beyond those we used in our experiments, we can further mention recent resource-aware optimizers [2, 8], evolutionary-based techniques [18, 34] and virtually any core improvement of BO, be it related to the acquisition function [48] or efficient parallelism [23, 47].”
- Deriving and improving CMA-ES with information geometric trust regions (2017, methodology): “CMA-ES uses a rank preserving transformation of fitness values which makes it invariant under monotone transformations of the fitness function [9].”
- Bayesian Optimization over Sets (2019, methodology): “CMA-ES [15] is applied in optimizing the acquisition function.”
- Bayesian optimization for conditional hyperparameter spaces (2017, methodology): “The population size used is λ = 4 + 3 log d, which is suggested as a rule of thumb in [15].”
- AutoFlow: Learning a Better Training Set for Optical Flow (2021, methodology): “To learn the hyperparameters for rendering the dataset, we develop a hybrid algorithm based on the populationbased training (PBT) algorithm [19, 4] and the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) algorithm [13].”
- Rethinking drug design in the artificial intelligence era (2019, methodology): “These latter approaches, typically in the class of estimation of distribution algorithms (EDA) 105 , such as Covariance Matrix Adaptation Evolutionary Strategies 106 , now overlap and are synergistic with machine learning methodology such as Information Geometric Optimization 107 and the Expectation”