Synergies between evolutionary algorithms and reinforcement learning

A recent trend in evolutionary algorithms (EAs) transfers expertise from and to other areas of machine learning. An interesting novel symbiosis considers: i) reinforcement learning (RL), which learns on-line and off-line difficult dynamic elaborated tasks requiring lots of computational resources, and ii) EAs with the main strength its eloquence and computational efficiency. These two techniques address the same problem of reward maximization in difficult environments that can include stochasticity. Sometimes, they exchange techniques in order to improve their theoretical and empirical efficiency, like computational speed for on-line learning, and robust behaviour for the off-line optimisation algorithms. For example, multi-objective RL uses tuples of rewards instead of a single reward value and techniques from multi-objective EAs should be integrated for an efficient exploration/exploitation trade-off. The problem of selecting the best genetic operator is similar to the problem an agent faces when choosing between alternatives in achieving its goal of maximising its cumulative expected reward. Practical approaches select the RL method that solve the best online operator selection problem.