中文摘要：

为降低深度学习程序的开发难度,提出了一种基于Julia云平台的交互式深度学习模式.设计了一套深度学习原语,用Julia实现原语,为Julia程序员提供调用接口,构建交互分析系统;并提供GPU/CPU实现接口,由云端系统根据用户情形自动优选;最后实现了基于深度学习的图像交互分析案例,验证原语体系的完备性和Julia云平台的交互支持能力.实验结果表明,GPU的运行速度比CPU快近30倍.Julia程序员只需通过调用原语,设置相关参数,就可以使用并行化的算法.

英文摘要：

In order to simplify the program development based on deep learning model, an in-teractive deep learning model based on Julia Cloud Platform is introduced in this paper. The con-tent of our study includes. Design a deep learning primitive function： Implement primitive func-tion in Julia language, supply interface for Julia programmer and construct interactive analysissystem. Implement both GPU and CPU interfaces, which can be automatically seIected by theCloud Platform according to whichever situation the clients are in. Implement image interactive a-nalysis case based on deep learning and verify the completeness of primitive function and interac-tive support ability of Julia Cloud Platform. The experiment result shows that running speed ofGPU case is up to 30 times faster than that of CPU case. Julia programmer only needs to callprimitive function and set some parameters to use the parallel algorithm.