A Reinforcement Learning Perspective on the Optimal Control of Mutation\n Probabilities for the (1+1) Evolutionary Algorithm: First Results on the\n OneMax Problem

2019/05/09 by Luca Mossina, Emmanuel Rachelson, Mossina, Luca +3
Computer Science · #Evolutionary Algorithms and Applications #Metaheuristic Optimization Algorithms Research #Reinforcement Learning in Robotics

paper · pdf · doi:10.48550/arxiv.1905.03726

Abstract

We study how Reinforcement Learning can be employed to optimally control\nparameters in evolutionary algorithms. We control the mutation probability of a\n(1+1) evolutionary algorithm on the OneMax function. This problem is modeled as\na Markov Decision Process and solved with Value Iteration via the known\ntransition probabilities. It is then solved via Q-Learning, a Reinforcement\nLearning algorithm, where the exact transition probabilities are not needed.\nThis approach also allows previous expert or empirical knowledge to be included\ninto learning. It opens new perspectives, both formally and computationally,\nfor the problem of parameter control in optimization.\n

Related

• Hybridizing the 1/5-th Success Rule with Q-Learning for Controlling the\n Mutation Rate of an Evolutionary Algorithm
• Multi-parameter Control for the (1+(λ,λ))-GA on OneMax via Deep Reinforcement Learning
• Probabilistic analysis of the (1+1)-evolutionary algorithm
• Control and Reinforcement Learning through the Lens of Optimization: An Algorithmic Perspective
• Reinforcement Learning and Control as Probabilistic Inference: Tutorial and Review