Contextual Covariance Matrix Adaptation Evolutionary Strategies

Authors: Abbas Abdolmaleki, Bob Price, Nuno Lau, Luis Paulo Reis, Gerhard Neumann

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental For illustration of the algorithm, we will use contextual standard functions and two contextual simulated robotic tasks which are robot table tennis, and a robot kick task. We show that our contextual CMA-ES algorithm performs favourably in comparison to other contextual learning algorithms.
Researcher Affiliation Collaboration Abbas Abdolmaleki1,2,3, Bob Price5, Nuno Lau1, Luis Paulo Reis2,3, Gerhard Neumann4 1: IEETA, DETI, University of Aveiro 2: DSI, University of Minho 3: LIACC, University of Porto, Portugal 4: CLAS, TU Darmstadt, Germany 5: PARC, A Xerox Company, USA
Pseudocode Yes Algorithm 1 Contextual Stochastic Search Algorithms
Open Source Code Yes Matlab code for reproducing the results on standard functions as well as videos regarding the experiments (table tennis and robot kick) at https://goo.gl/MLz Ks W
Open Datasets Yes We chose two series of optimization tasks for comparisons. In the first series, we use the standard optimization test functions [Molga and Smutnicki, 2005]
Dataset Splits No No explicit training/validation/test dataset splits (percentages, counts, or references to predefined splits) were found. The paper describes an iterative optimization process based on generated samples rather than fixed data partitions.
Hardware Specification No No specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running experiments were mentioned.
Software Dependencies No For example we use fmincon tool in matlab.
Experiment Setup Yes For all experiments, the KL-bound ϵ for REPS is set to 1 and for all experiments we use the default hyper parameter settings given in Algorithm 1 without further tuning.