The Cross-Entropy Method for Combinatorial and Continuous Optimization

By Reuven Y. Rubinstein. 1999

How researchers see this paper

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

In Optimization & Training

Citations in the corpus

Benchmarking Deep Reinforcement Learning for Continuous Control (2016, methodology): “We also observe that CEM outperforms CMA-ES, which is remarkable as CMA-ES estimates the full covariance matrix.”
How to train your MAML (2018, methodology): “In this paper we use the cross-entropy (De Boer et al. (2005); Rubinstein (1999)) loss throughout.”
Composable Action-Conditioned Predictors: Flexible Off-Policy Learning for Robot Navigation (2018, methodology): “We use stochastic optimization based on the cross entropy method [31] to solve this maximization, although in principle any optimizer could be used.”
SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation (2020, methodology): “In this category we benchmark the cross entropy method (CEM) [48] as a representative random shooting algorithm.”
code2seq: Generating Sequences from Structured Representations of Code (2018, methodology): “We optimize the cross-entropy loss (Rubinstein, 1999; 2001) with a Nesterov momentum (Nesterov, 1983) of 0 .”
Conditioning by adaptive sampling for robust design (2019, methodology): “• In CEM-PI , w ( x i ) = 1 { PI ( X ) ≥ β ( t ) } ( x i ) where PI ( x ) is the probability of improvement function and β ( t ) is adjusted according to the methods of CEM (there is a parameter in CEM that corresponds to our Q, which we set to 0.8 for CEM-PI ).”
Up or Down? Adaptive Rounding for Post-Training Quantization (2020, methodology): “We use the cross-entropy method (Rubinstein, 1999) to solve the QUBO problems in (13) and (20), where we initialize the sampling distribution for the binary random variables ŵi as in (Gupta et al.”
HELoC: Hierarchical Contrastive Learning of Source Code Representation (2022, methodology): “NEP employs a cross-entropy loss function [39], denoted as Lh , and NRO employs a triplet loss function [40], denoted asLt .”
Ant Colony Optimization (2001, unknown): “, 1997), cross-entropy (Rubinstein, 1999), stochastic gradient descent (Robbins & Monroe, 1951), and estimation of distribution algorithms (Larrañaga & Lozano, 2001).”
Sim2Real View Invariant Visual Servoing by Recurrent Control (2017, methodology): “In our implementation, we use the cross-entropy method (CEM) [28] to perform this optimization, which provides a simple gradient-free optimization procedure for choosing actions that maximize predicted Q-values.”
HEAVY TAILS, IMPORTANCE SAMPLING AND CROSS–ENTROPY (2005, methodology): “3 The cross–entropy method The cross-entropy method originated from an adaptive method for estimating probabilities of rare events in complex stochastic networks [16], and has quickly evolved into a versatile and unified method for efficient simulation and combinatorial and multi-extremal continuous”
Sim2Real Viewpoint Invariant Visual Servoing by Recurrent Control (2018, methodology): “We then use CEM to perturb this action with Gaussian noise with standard deviation σ = 0.003 to generate 150 candidate actions and evaluate them via the value prediction head.”
Sequence Modeling of Temporal Credit Assignment for Episodic Reinforcement Learning (2019, methodology): “Besides policy optimization methods, blackbox optimization approaches , such as cross entropy method [29], CMA-ES [11] and evolution strategies [30], have been also applied to episodic RL problems due to their computational efﬁciency.”
Policy Optimization via Importance Sampling (2018, methodology): “In contrast, in parameter-based methods, the search is carried out directly in the space of parameters by exploiting global optimizers [e.g., 40, 15, 47, 51] or following a proper gradient direction like in Policy Gradients with Parameter-based Exploration (PGPE) [44, 62, 45].”
Chapter 3 – The Cross-Entropy Method for Optimization (2013, background): “Subsequent work by Rubinstein (1999; 2001) has shown that many optimization problems can be translated into a rare-event estimation problem.”
Sample-Efficient Optimization in the Latent Space of Deep Generative Models via Weighted Retraining (2020, methodology): “The cross-entropy (CE) method iteratively retrains a generative model using a weighted training set, such that high-scoring points receive higher weights [57, 56, 12].”
DeepType: Multilingual Entity Linking by Neural Type System Evolution (2018, methodology): “Cross-Entropy Method (CEM) (Rubinstein 1999) is a stochastic optimization procedure applicable to the selection”