智能部署的可扩展AI框架研究

Research on Expandable AI Framework For Intelligent Deployment

开源人工智能框架针对典型问题能够直接调用固有算法并快速建模,但在特定业务场景中存在研究、实现和部署等环节协同处理效率不高的问题。因此,本文提出了一种智能部署的可拓展人工智能框架方案。首先,采用数据框实现不同算法间(包括单机和分布式间)的转化;其次,提出可配置的算法管道模型,实现算法的流水线运转和可扩展性;最后,基于虚拟化技术实现智能部署。实际使用结果表明,该方案可提高各环节协同处理时间效率约20%,并具有良好的可实施性。

The open source AI framework could directly call the inherent algorithms and quickly model typical problems,but there was a problem of low efficiency in the collaborative processing of research,implementation and deployment in specific business scenarios,an expandable AI framework for intelligent deployment has been studied.Firstly,Data frame structure was used to realize the transformation between different algorithms(including stand-alone and distributed);Secondly,a configurable algorithm pipeline model was proposed to realize the pipeline operation and scalability of AI algorithm;Finally,the intelligent deployment of AI framework was realized based on virtualization technology.The actual application results show that the AI framework could improve the collaborative processing efficiency of all links about 20%and has good feasibility.