Research And Implementation Of Visual Programming System For Deep Learning Applications

In recent years,with the development and advancement of Machine Learning technology,especially the successful application of Deep Learning algorithms in the field of image and big data processing,how to efficiently develop and apply Machine Learning technology has become a hot issue.At present,the development method of Deep Learning algorithm is based on text programming language.There are disadvantages such as high learning threshold and low development efficiency.A tool or platform is needed to reduce the learning threshold for Deep Learning modeling and to present the intrinsic structure of the model in an easy-tounderstand manner.This paper designs and implements a Visual Programming System for Deep Learning modeling.Compared with the text programming method,Visual Programming uses the graphical representation algorithm and dataflow,without complicated code engineering,and easily grasp the algorithm construction and data processing.method.This paper introduces and summarizes the research status of Visual Programming and the development of Machine Learning visualization,analyzes the processing and characteristics of constructing Deep Learning model.For the Deep Learning modeling and application using Visual Programming methods,the main work of this paper is as follows:(1)Design a Visual Programming Language based on dataflow.Based on the Dataflow Computational Graph model representing the algorithm calculation process.According to the actual needs of deep learning modeling,design the visual element definition,formal description and functional specification of the Visual Programming Language.(2)Design and improved the code generation algorithm of the Visual Programming Language.Optimize the Dataflow Computational Graph based on dependency scan and hierarchical topology sorting.An edge-based Dataflow Computational Graph code generation algorithm is proposed to generate Python code from the bottom up by defining a property map rewrite rule.(3)Design and implement a Visual Programming System.Ability to cross-platform access and visual programming in a WYSIWYG manner,supporting undo/redo functionality.Graphically display the logic of the Dataflow Computational Graph,and interactively construct the Deep Learning algorithm model.Generate Deep Learning model and code file based on MXNet framework.