2019/05/20 by Mudita Sharma, Sharma, Mudita, Alexandros Komninos +5 · 3 citations
Computer Science · #Evolutionary Algorithms and Applications #Metaheuristic Optimization Algorithms Research #Reinforcement Learning in Robotics
paper · pdf · doi:10.48550/arxiv.1905.08006
Adaptive Operator Selection (AOS) is an approach that controls discrete\nparameters of an Evolutionary Algorithm (EA) during the run. In this paper, we\npropose an AOS method based on Double Deep Q-Learning (DDQN), a Deep\nReinforcement Learning method, to control the mutation strategies of\nDifferential Evolution (DE). The application of DDQN to DE requires two phases.\nFirst, a neural network is trained offline by collecting data about the DE\nstate and the benefit (reward) of applying each mutation strategy during\nmultiple runs of DE tackling benchmark functions. We define the DE state as the\ncombination of 99 different features and we analyze three alternative reward\nfunctions. Second, when DDQN is applied as a parameter controller within DE to\na different test set of benchmark functions, DDQN uses the trained neural\nnetwork to predict which mutation strategy should be applied to each parent at\neach generation according to the DE state. Benchmark functions for training and\ntesting are taken from the CEC2005 benchmark with dimensions 10 and 30. We\ncompare the results of the proposed DE-DDQN algorithm to several baseline DE\nalgorithms using no online selection, random selection and other AOS methods,\nand also to the two winners of the CEC2005 competition. The results show that\nDE-DDQN outperforms the non-adaptive methods for all functions in the test set;\nwhile its results are comparable with the last two algorithms.\n