The Cross‐Entropy Method

By Reuven Y. Rubinstein, Dirk P. Kroese. 2004

How researchers see this paper

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

Citations in the corpus

QT-Opt: Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulation (2018, methodology): “We use the crossentropy method (CEM) to perform this optimization, which is easy to parallelize and moderately robust to local optima for low-dimensional problems [36].”
Learning hand-eye coordination for robotic grasping with deep learning and large-scale data collection (2016, methodology): “However, we can obtain better results by running a small optimization on vt, which we perform using the cross-entropy method (CEM) (Rubinstein and Kroese, 2004).”
Quantile QT-Opt for Risk-Aware Vision-Based Robotic Grasping (2019, methodology): “The cross-entropy method (CEM) [22] is used to iteratively optimize and select the best action for a given Q-function:”
Cross-Entropy Optimization for Independent Process Analysis (2006, unknown): “In the present work, θ contains transition probabilities i → j (1 ≤ i, j ≤ D), called node transition parametrization in the literature [17].”
Entity Abstraction in Visual Model-Based Reinforcement Learning (2019, methodology): “The actions are optimized using the cross-entropy method (CEM) [59], with each sampled action evaluated by the greedy cost function described in Sec.”
Time-Agnostic Prediction: Predicting Predictable Video Frames (2018, methodology): “To generate the data, we use a cross-entropy method (CEM)-based planner (Kroese et al., 2013; Ebert et al., 2017) in the MuJoCo (Todorov et al., 2012) environment with access to the physics engine, which produces non-deterministic trajectories.”
A Sequential Importance Sampling Algorithm for Generating Random Graphs with Prescribed Degrees (2011, methodology): “Important variations which could be coupled with the present algorithms include Owen’s [60] use of control variates to bound the variance and adaptive importance sampling as in Rubinstein [64].”
Adaptive methods for sequential importance sampling with application to state space models (2008, methodology): “(5.2) Since the expectation in (5.2) is intractable in most cases, the key idea is to approximate it iteratively using IS from more and more accurate approximations—this idea has been successfully used in CE methods; see e.g. Rubinstein and Kroese (2004).”
The Generalized Cross Entropy Method, with Applications to Probability Density Estimation (2011, methodology): “Our approach was motivated by the work of Kapur and Kesavan [30] and the well-known Cross Entropy (CE) method [50, 49], and will be referred to as the Generalized Cross Entropy method (GCE) due to its emphasis on using generalized CE measures, as opposed to using the traditional Shannon and Kullback”
Basis Function Adaptation in Temporal Difference Reinforcement Learning (2005, methodology): “We consider here two algorithms for adjusting the basis function parameters: The first is a local, gradient-based method, while the second is based on the Cross Entropy (CE) method for global optimization (Rubinstein, 1999; Rubinstein and Kroese, 2004).”
Sequential design of computer experiments for the estimation of a probability of failure (2010, methodology): “…a host of refined random sampling methods have been proposed to improve over the basic Monte Carlo convergence rate; for instance, methods based on importance sampling with cross-entropy (Rubinstein and Kroese, 2004), subset sampling (Au and Beck, 2001) or line sampling (Pradlwarter et al., 2007).”
Minimum variance importance sampling via Population Monte Carlo (2007, methodology): “Note however that a recent reference is the cross-entropy method of [21] where the parameter of a family of proposals is optimized, either directly or by an iterative process, to reach minimal variance or maximal entropy against the target |h(x)|π(x), the function h being of the specific rare event ”
Improvisation through Physical Understanding: Using Novel Objects as Tools with Visual Foresight (2019, methodology): “In practice, we choose the number of samples M to be 100, the horizon H to be 21, and the number of CEM iterations n iter to be 3.”
Adaptive sequential Monte Carlo by means of mixture of experts (2011, methodology): “…followed by resampling) can be viewed as standard importance sampling where a weighted mixture of N stratas—the i th stratum being proportional to ω i l( X i , · ) — approximating ˜ μ is targeted using a mixture proposal π comprising N stratas proportional to { ω i r( X i , · ) } Ni = 1 .”
Adaptive importance sampling in general mixture classes (2007, methodology): “In the sequel, we refer to the criterion in (2) as the entropy criterion since it is obviously related to the performance measure used in the cross-entropy method of Rubinstein and Kroese (2004).”
Joint additive Kullback–Leibler residual minimization and regularization for linear inverse problems (2007, background): “The book [15] examines the properties of cross-entropy in considerable detail by, among other things, exploiting its expectation interpretation.”