Sophon base 3.1 launches mlops function to provide wings for enterprise AI capability operation

Guided by domestic policy 、 Under the influence of multiple factors such as the original demand and epidemic situation of enterprise industrial intelligence upgrading , The industrial application of artificial intelligence in China has been in the past 5 It has shown an unparalleled growth rate over the years . However, with the deepening of the application of artificial intelligence and machine learning in all business levels of enterprises , The type and quantity of models are also growing rapidly . Due to the training framework between different models 、 Deployment mode 、 Input and output are different , This makes it difficult for heterogeneous models to be managed uniformly . meanwhile , There is a lack of unified model and standardized process specification from development to launch among departments , As a result, model assets are scattered in various business departments , It is difficult to be uniformly managed as enterprise data assets . Increasing model management costs and O & M complexity , And uncontrollable risks in the production environment , It brings a series of challenges to enterprise model management .

To solve the problem AI It's hard to land , Star Ring Technology AI The team starts from the user demand side , We have made great efforts to develop an enterprise platform based on cloud native architecture AI Capability operation platform Sophon MLOps, boost AI Model to the ground .

Sophon MLOps  It is an enterprise based on cloud native architecture AI Capability operation platform , Focus on model management in the whole life cycle of machine learning model 、 Model deployment 、 Model monitoring and early warning 、 Key links such as model evaluation and model iteration .

Through unified tube 、 Unified operation and maintenance 、 Unified application 、 Unified monitoring 、 Unified evaluation 、 Unified interpretation , Give enterprise customers ease of use 、 Efficient, safe and reliable AI Ability to operate services , Assist customers in large-scale management of growing machine learning models , Improve model efficiency , Reduce the cost of model integration management , Control model production environment risk .

Sophon MLOps For businesses AI The pain point of operation , Around the enterprise AI Model access 、 operation management 、 The whole life cycle of continuous training , Provide large-scale integrated management 、 Efficient model reasoning 、 Model monitoring and early warning 、 Model performance evaluation 、 Privacy security and other functions , For the enterprise AI Daily operation with wings .

Star Ring Technology ​findermp.video.qq.com/251/20304/stodownload?encfilekey=rjD5jyTuFrIpZ2ibE8T7YmwgiahniaXswqztOl5wQZqIRKPoEqwq1HLN7SudcdFQ691icRlSzbBt6MgTZXccy46hagnu4InGcjDoze2X8YI7UVJIMEzSmmqdcA&adaptivelytrans=0&bizid=1023&dotrans=0&hy=SH&idx=1&m=&scene=0&token=AxricY7RBHdUzqcYx8FhNSYQRds360HTKgic22q50gQ86TppL4E2KjznT7z3xYoAjjGcuTKs382sc

This update ,MLOps To upgrade to v1.1 edition , Add the following new functions to the existing functions ：

• New model service
  Batch Forecasting
  function ： Improved the application scenario , The function coverage range is from the original online real-time prediction , Upgrade to support both online and offline modes . The offline batch forecasting function supports the direct connection to the database , Obtain batch data for model prediction , And write the results back to the corresponding library table , Achieve closed loop ;

• New model service
  Data migration
  Indicator monitoring function ： Strengthened the ability of model service in monitoring and early warning , It is committed to helping users more comprehensively master the running status of machine learning model services , And find and handle abnormal conditions in time through self-defined monitoring indicators , Avoid the risk caused by data offset and other reasons ;

• New model service monitoring indicator configuration and
  The alarm
  function ;

• New service release
  Approval process
  management ;

• newly added
  Interpretable machine learning XAI modular
  (MVP edition )： It can handle the correlation interpretation in the process of data analysis 、 Procedural interpretation 、 Inferential interpretation and causal interpretation . Users can finely analyze the influence relationship between features and results , Help users accurately improve DataCentric-AI Data governance capabilities , Targeted optimization of model accuracy , Help users quickly locate 、 Optimize key factors that impact business results , To drive business success .

In the financial technology industry , As regulatory policy continues to tighten , The China Banking and Insurance Regulatory Commission on 2020 year 7 In June, it was officially introduced 《 Measures for the administration of Internet loans of commercial banks 》, Commercial banks are required to implement the model from development to testing 、 review 、 Risk management throughout the life cycle of exit monitoring . On the one hand, in order to meet the regulatory compliance requirements , On the other hand, improve the management efficiency of intra bank risk model , The bank requires unified management of the whole life cycle of the model . Besides , With the continuous development of intra bank business , Massive isomerism AI Model assets are scattered across departments , Once needed , The deployment cycle of the model is prolonged due to the participation and inter departmental coordination .

Sophon Base 3.1 Use MLOps The bank has established a unified AI Model management platform , Fast access to a large number of model files generated by the training of different frameworks or platforms accumulated in the bank , Manage model assets by version ; And establish a standardized process , Unified way to build model reasoning logic , Support zero code one click deployment model application . Based on cloud native infrastructure , Get through the whole life cycle process of the model , It realizes the unified operation, maintenance and monitoring of the model application by the bank .

After the platform goes online ,Sophon Base Hundreds of machine learning models generated by various algorithm frameworks of the bank are integrated . The average time to deploy the model application is determined by 1.5 The sky falls to 0.5 Hours , The configuration cost is reduced by nearly 80%, The average iteration period of the model is changed from 1 Monthly decrease 1 Zhou . In terms of model efficiency , Support hundreds of model prediction services online at the same time , A single piece of data realizes millisecond response .

Sophon MLOps Opened the AI The whole life cycle of , It has built a unified platform for all kinds of user roles in the enterprise AI Collaboration platform . For enterprises ,MLOps Large scale integrated management of multi-source heterogeneous machine learning model , And provide an efficient and secure model reasoning 、 Monitoring, early warning and performance evaluation services ; For users , Feel the speed of operation ,AI Application and deployment are even more powerful . future ,MLOps We will continue to iterate over richer features , Empowering enterprise AI faster 、 Better landing .