Please scan the QR code from your WeChat application:

Updates
These companies currently have access to the the MLOps to improve lowcode based AI Catalyst. Would you like the following company to keep you updated about their products and services by electronic means?
Beijing Baidu Netcom Science Technology Co., Ltd.
Beijing Tianyuan DIC Information Technology Co. Ltd.
Business-intelligence of Oriental Nations Corporation Ltd.
China Telecommunications Corporation
China Unicom
Description:

Please confirm that you would like to stop following this catalyst project.

 

Confirm Cancel

Log in

 

Not a user?

Register

0 Review(s)
0
Average based on 0 review(s)
5
0
4
0
3
0
2
0
1
0
URN
C22.0.371
Themes
AI , Data & Insights , Cloud Native IT & Networks
Topics
Artificial Intelligence (AI)

What is the main problem identified?

Telcos collect a massive amount of data for analytics and insights. How we can reduce the cost of data storage and transition? 
Group data is stored in various business systems in different forms, and the data storage cost is high. Before using the data, it needs to go through multiple processes such as data application, data preprocessing and feature processing. The process is lengthy, repetitive and inefficient, which is not conducive to the rapid production of the model.

Capability to drive insight from data require specific knowledge and competence, thus difficult to scale in a short period of time, is there a lowcode solution to lower the entry barrier? Based on the massive business data of China Unicom, large-scale distributed model training and reasoning are needed, which requires high skills of model developers. It takes a long time of knowledge accumulation and practical activities to be competent for model development.

The algorithm/ model based on the lowcode approach, usually yields poorer AI quality, if MLops can be applied to improve the result.

Who is affected most by this problem?

Head of the digital department of Telco

How will this project team solve the problem?

In this Catalyst, our plan is:
1) Establishing a unified data source and feature management function
the algorithm model platform connects all business systems uniformly, collects all business data, and establishes a unified data source and feature management center. Model developers only need to apply to use data and features in the platform, which can greatly reduce the time from model development to application deployment and promote the rapid iterative development of business.
2) Establishing a visual, low code model development model
The algorithm model platform provides users with two model development methods: visual modeling and coding modeling, which can meet the diversified needs of users and realize rapid modeling.
3)  If we are able to improve the quality of selected algorithms/models by deploy an MLops infra

How will the success of the solution be measured?

1)    Evaluating indicator of establishing a unified data source and feature management function
Reducing the amount of data storage and transmission through the data source and feature management center;
Realizing feature processing support, consumption support, and monitoring support;
Realizing data audit such as data validity, consistency, and integrity to ensure data quality;
Unified quality audit and authority control of data sets and features.

2)    Evaluating indicator of establishing a visual and low code model development mode
Provide low code and modeling by dragging and dropping;
The platform supports mainstream distributed machine learning and deep learning algorithm frameworks, such as TensorFlow, PyTorch, spark, etc;
Providing visual modeling workbench and convenient modeling method by dragging and dropping, realizing low code development mode, and help users quickly realize large-scale distributed model training and reasoning;
The platform provides unified computing resource allocation and monitoring, elastic shrinkage of memory, etc., making effective use of resources.

3) Deploy an MLops infra to improve the selected algorithm/models to see if we are able to improve the quality of them.