An artificial intelligence platform system based on deep learning

Abstract

The invention discloses an artificial intelligence platform system based on deep learning, comprising: a platform layer for authority management, distributed storage, GPU computing resource management, distributed computing and training, and task scheduling; a model layer for providing machine Learning model and deep learning model; application layer, used for resource management and monitoring, model definition and training, providing interactive transformation environment, intelligent data labeling and model export and release. The present invention develops an AI platform system through engineering means to improve the utilization rate of GPU and other hardware resources, reduce hardware investment costs, and help algorithm engineers to apply various deep learning technologies more conveniently, so that it can avoid complicated environment operation and maintenance. Freed up, it provides efficient storage of massive training data, isolation of user resources, and more secure access control; unified management of training data and training tasks, standardized and streamlined machine learning processes; automated data labeling to improve model iteration efficiency.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, in particular to an artificial intelligence platform system based on deep learning. Background technique [0002] In recent years, with the rapid development of artificial intelligence technology, deep learning has swept every corner of IT, changing the model of algorithm research and software development in various fields, and also bringing great benefits to IT infrastructure construction and platform tool development. Here comes a new requirement. Quickly build a distributed deep learning training platform and accelerate the training of deep neural networks, which can effectively improve the company's competitiveness. [0003] Currently, there are many deep learning frameworks, such as mxnet, tensorflow, cntk, etc., which are familiar to everyone. These frameworks are developed in different languages ​​and have different interface designs. many difficulties. [0004] One of t...

AI Technical Summary

Problems solved by technology

[0007] 1. It does not support GPU resource scheduling management. Under the load of deep learning, GPU is a first-class citizen of resource scheduling, and GPU resources cannot be used, which has a great impact on the efficiency of model training;

[0008] 2. It only supports data storage, lacks support for data labeling, and needs to seek additional labeling tools;

[0009] 3. There is no authority control system, and data security cannot be guaranteed;

[0010] 4. There is no interactive environment, and the development efficiency of algorithm engineers is low

Images