biaohe / den-armoea Goto Github PK

This project forked from xw00616/den-armoea

# Introduction of DNN-AR-MOEA This repository contains code necessary to reproduce the experiments presented in Evolutionary Optimization of High-DimensionalMulti- and Many-Objective Expensive ProblemsAssisted by a Dropout Neural Network. Gaussian processes are widely used in surrogate-assisted evolutionary optimization of expensive problems. We propose a computationally efficient dropout neural network (EDN) to replace the Gaussian process and a new model management strategy to achieve a good balance between convergence and diversity for assisting evolutionary algorithms to solve high-dimensional multi- and many-objective expensive optimization problems. mainlydue to the ability to provide a confidence level of their outputs,making it possible to adopt principled surrogate managementmethods such as the acquisition function used in Bayesian opti-mization. Unfortunately, Gaussian processes become less practi-cal for high-dimensional multi- and many-objective optimizationas their computational complexity is cubic in the number oftraining samples. # References If you found DNN-AR-MOEA useful, we would be grateful if you cite the following reference: Evolutionary Optimization of High-DimensionalMulti- and Many-Objective Expensive ProblemsAssisted by a Dropout Neural Network (IEEE Transactions on Systems, Man and Cybernetics: Systems).