Cross-platform drag-and-drop deep learning modeling and training method and device

Abstract

The invention discloses a cross-platform drag-and-drop deep learning modeling and training method and device, wherein the method includes: collecting training data according to the target application scene; obtaining preprocessed images through online annotation, and dragging different algorithm modules Combining training models to generate an initial solution model; training the initial solution model according to a training request to obtain a final solution model and displaying a training result. According to the method of the embodiment of the present invention, a solution model can be generated through comprehensive modeling aimed at the overall solution, which is simple and easy to implement.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a cross-platform drag-and-drop deep learning modeling and training method. Background technique [0002] In related technologies, the visualization software platform includes data preprocessing, feature extraction to realize model training and other functions. Specifically, based on stand-alone or C / S architecture, C++, JavaScript, and Java programming languages ​​are used for operations, and simple scripting languages ​​can also be used to automatically perform process operations, and in the interface, users can drag and drop some graphical algorithm components In the visual interface, establish the data flow between the graphical algorithms, and then perform model training. In related technologies, machine learning in visualization tools generally achieves model training by extracting features of raw data and selecting an appropriate algorithm. [0003] However, ther...

AI Technical Summary

Problems solved by technology

[0005] 2. Once it is based on the C / S architecture, you need to download the specified client application, which has a long update cycle and high cross-platform costs;

[0006] 3. Use specified and commonly used machine learning algorithms, but the compatibility of independent algorithms for specific needs is not strong, and the algorithm scalability is not good;

[0007] 4. Many software client interface descriptions use professional terms, which cannot be easily used by general analysts, but require a long learning period;

[0008] 5. The data source of the software platform is mainly the mainstream big data analysis platform, but there are relatively large differences in the compatibility of multiple software clients, which leads to filtering, deduplication, splitting, and merging of data to achieve data visualization. , and realize the difficulty in the operation of feature extraction, management and reuse of data

Images