.png){kind=icon}
.png){kind=icon}
Return
Deep Deterministic Evolutionary Algorithm Fusing Genetic Evolution With Deep Reinforcement Learning
.png){kind=icon}
PRE
AI
DOI：10.4018/IJCINI.407549.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep reinforcement learning has demonstrated significant potential in solving complex control tasks, yet it remains constrained by persistent challenges because of hyperparameter sensitivity. However, conventional evolutionary algorithms often exhibit high sample complexity and struggle with optimizing high-dimensional parameter spaces. The authors propose a deep deterministic evolutionary algorithm that fuses genetic evolution with deep reinforcement learning. The proposed algorithm leverages populations to generate diverse exploration data and periodically infuses gradient-based updates from a deep deterministic policy gradient agent into the evolving population. Extensive experiments that focused on challenging multi-joint dynamics with contact continuous control benchmarks demonstrated that the proposed algorithm outperforms state-of-the-art deep reinforcement learning methods and stand-alone genetic algorithm. Key results highlight superior sample efficiency, stable convergence, and robustness to deceptive local optima, in particular in environments with sparse rewards.
Keywords:
Machine Learning
Deep Learning
Reinforcement Learning
Evolutionary Algorithms
Policy Gradients
Journal
I
IF:
1
Papers:
14
Citations:
137