Completely Derandomized Self-Adaptation in Evolution Strategies

By Nikolaus Hansen, Andreas Ostermeier. 2001

4,706 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Optimization & Training

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

In Optimization & Training

Citations in the corpus

Benchmarking Deep Reinforcement Learning for Continuous Control (2016, methodology): “For higher-dimensional policy parameterizations, the computational complexity and memory requirement for CMA-ES become noticeable.”
Recurrent World Models Facilitate Policy Evolution (2018, methodology): “To optimize the parameters of C, we chose the Covariance-Matrix Adaptation Evolution Strategy (CMA-ES) [29, 30] as our optimization algorithm since it is known to work well for solution spaces of up to a few thousand parameters.”
Deep learning in neural networks: An overview (2014, methodology): “…& Schmidhuber, 1997; Saravanan & Fogel, 1995), Covariance Matrix Estimation Evolution Strategies (CMA-ES) (Hansen, Müller, & Koumoutsakos, 2003; Hansen & Ostermeier, 2001; HeidrichMeisner & Igel, 2009; Igel, 2003), and NeuroEvolution of Augmenting Topologies (NEAT) (Stanley & Miikkulainen,…”
Taking the Human Out of the Loop: A Review of Bayesian Optimization (2016, methodology): “[158] use the CMA-ES method of [66], and Hutter et al.”
World Models (2018, methodology): “To optimize the parameters of C, we chose the Covariance-Matrix Adaptation Evolution Strategy (CMA-ES) (Hansen, 2016; Hansen & Ostermeier, 2001) as our optimization algorithm since it is known to work well for solution spaces of up to a few thousand parameters.”
Active learning of inverse models with intrinsically motivated goal exploration in robots (2013, neutral): “local optimization. 3.6.2 Exploration Phase This phase consists in performing q2N small random explorative actions i, around the current position , where the variations can be derandomized such as in [45]. This allows the learning system to improve its regression model of the relationship (; ) ! y, ”
Google Vizier: A Service for Black-Box Optimization (2017, methodology): “In Figures 6 we look at result quality for four optimization algorithms currently implemented in the Vizier framework: a multiarmed bandit technique using a Gaussian process regressor [29], the SMAC algorithm [19], the Covariance Matrix Adaption Evolution Strategy (CMA-ES) [16], and a probabilistic ”
One Pixel Attack for Fooling Deep Neural Networks (2017, methodology): “elongs to a big class of algorithms called evolutionary strategies [48] which includes other variants such as Adaptive DE [49] and Covariance matrix adaptation evolution strategy (CMA-ES) [50], [51], [52]. In fact, there are a couple of recent developments [61], [62], [63] in evolutionary strategies”
Learning Visual Predictive Models of Physics for Playing Billiards (2015, methodology): “In practice, instead of exhaustively searching for all forces we use CMA-ES method (Hansen & Ostermeier, 2001) for determining the optimal force.”
Evolution Strategies as a Scalable Alternative to Reinforcement Learning (2017, background): “The most widely known member of the ES class is the covariance matrix adaptation evolution strategy [CMA-ES; Hansen and Ostermeier, 2001], which represents the population by a full-covariance multivariate Gaussian.”
On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models (2015, methodology): “1, except that C is now an FNN or RNN trained by evolutionary algorithms [200, 255, 105, 56, 68] applied to NNs [165, 321, 180, 259, 72, 90, 89, 110, 94], or by policy gradient methods [314, 315, 316, 274, 18, 1, 63, 128, 313, 210, 192, 191, 256, 85, 312, 190, 82, 93][245, Sec.”
Entropy Search for Information-Efficient Global Optimization (2011, methodology): “The expected loss is (3) where L ( p min ( x | y , x )) should be understood as the cost assigned to the measure p min ( x ) induced by the posterior belief over f after observations y = { y 1 , . . . , y H } ⊂ R at the locations x .”
Model-Free Preference-Based Reinforcement Learning (2016, methodology): “The GPs hyperparameter are automatically tuned, using a CMA-ES (Hansen and Ostermeier 2001) after each iteration.”
Self-Delimiting Neural Networks (2012, neutral): “A more general approach for partially observable environments directly evolves programs for RNNs with internal states (no need for the Markovian assumption), by applying evolutionary algorithms [45, 68, 28] to RNN weight matrices [82, 70, 72, 25].”
Online Batch Selection for Faster Training of Neural Networks (2015, methodology): “We used Covariance Matrix Adaptation Evolution Strategy (CMA-ES) by Hansen & Oster-meier (2001) in its “active” regime (Hansen & Ros, 2010) as implemented in a Python code available at https://pypi.python.org/pypi/cma .”
CMA-ES for Hyperparameter Optimization of Deep Neural Networks (2016, neutral): “The Covariance Matrix Adaptation Evolution Strategy (CMA-ES (Hansen & Ostermeier, 2001)) is a state-of-the-art optimizer for continuous black-box functions.”
Path Integral Policy Improvement with Covariance Matrix Adaptation (2012, methodology): “For a full explanation of the algo-rithm we refer to Hansen & Ostermeier (2001).”