EvoX: A Distributed GPU-accelerated Framework for Scalable Evolutionary Computation

Evolutionary Computation (EC), drawing inspiration from natural evolutionary processes, has solidified its place as an integral facet of Artificial Intelligence. Its unique attributes, such as adaptability and the capability to navigate vast problem spaces, have rendered it indispensable, especially in domains demanding optimization like engineering design. In today's data-driven landscape, the need for scalability in EC is more pronounced than ever, especially with the rise in complex systems and large-scale data. However, many existing EC libraries, designed for modest scales, fall short in catering to the heightened demands of modern problems. The advent of some pioneering GPU-accelerated EC libraries is a step forward, but they too grapple with limitations, particularly in terms of flexibility, computational efficiency, and architectural robustness. To address these challenges, this paper introduces EvoX: a comprehensive, scalable framework tailored for the automated, distributed, and heterogeneous execution of EC algorithms. Central to EvoX is a functional programming model that streamlines the EC algorithm development process, bolstered by a hierarchical state management strategy for efficient distributed execution. Alongside this, leveraging the capabilities of EvoX, we present a rich library of EC algorithms designed to handle a spectrum of problem-solving scenarios. Experimental results demonstrate both the superior system performance and model performance of EvoX. The code of EvoX is available at https://github.com/EMI-Group/EvoX