Search

Reinforcement Learning (RL) has gained significant popularity in recent years for its ability to solve complex control problems. However, most existing RL algorithms are designed to train policies for each environment in isolation, limiting their applicability to real-world scenarios with many related environments. Recently, many multitask optimization algorithms have been proposed and successfully applied to a wide range of optimization problems. However, existing studies mainly focus on solving multiple continuous function benchmarks and overlook the potential of tailoring towards RL. In this paper, we propose a simple multitask optimization algorithm called Multitask Augmented Random Search (MARS) that trains multiple RL agents together and exploits the performance surplus from highly correlated tasks. MARS is a modification of the simple random search Augmented Random Search (ARS) algorithm, which has been shown to outperform complicated methods in solving continuous control MuJoCo environments such as Soft Actor-Critic (SAC), Proximal Policy Optimization (PPO), and Trusted Region Policy Optimization (TRPO). The experimental results also demonstrate that our proposed algorithm is more consistent in solving different instances of MuJoCo benchmark than ARS, Multifactorial Evolution Algorithm (MFEA), and Adaptive MFEA RL (AMFEARL) within the same number of training episodes. • Propose a multitask extension of ARS for training related continuous control problems. • Derived an online data-driven measurement to preempt the transfer of solutions incomplementary. • Conduct experiments on synthetic benchmarks and in the multitask Mujoco environments. [ABSTRACT FROM AUTHOR]

The differential evolution (DE) algorithm is widely regarded as one of the most influential evolutionary algorithms for addressing complex optimization problems. However, the fixed mutation strategy limits the adaptive ability of DE, and the lack of utilization of historical information limits the optimization ability of DE. In this paper, an indicator-based self-learning differential evolution algorithm (ISDE) is proposed. A jump out mechanism based on deep reinforcement learning is adopted to control the mutation intensity of the population. The neural network in the jump out mechanism is designed as a decision maker. The mutation intensity of the population is controlled by the neural network, and the neural network are trained by a double deep Q network algorithm based on the continuous data generated during the evolution process. A population range indicator (PRI) is utilized to describe individual differences in the population. A diversity maintenance mechanism is designed to maintain individual differences according to the value of PRI. The experimental results reveal that the comprehensive performance of ISDE is superior to comparison algorithms on CEC 2017 real-parameter numerical optimization. [ABSTRACT FROM AUTHOR]