In the near future, AI is expected to see extensive use in various industries and businesses. We are in the midst of what has been called "the era of AI," and hopes for AI are rising. However, facilitating the use of AI involves numerous challenges, among them the effort and expense involved in building systems and the impact that environmental changes have on systems that have already been put into operation. To tackle these challenges, we developed the MLOps Platform using the technologies and knowledge the Toshiba Group has accrued through its experience with researching and deploying AI. Let's learn a little about this MLOps Platform. 

The Challenges Involved in Building AI-based Systems

 Modern AI has been developed through long years of technological innovation, and it is now ready for extensive use in business.

 For example, in plants with numerous IoT devices, the data collected from devices can be analyzed by AI to predict failures with a high degree of accuracy. Some companies are analyzing the massive amounts of data collected through their electronic commerce (EC) sites and using the results in marketing, such as by providing customers with recommendations or predicting product demand.

These diverse analyses are performed using AI models. AI models make future predictions using input data categories and previously accrued data. AI model processing is generated through machine learning based on large amounts of data.

Using high quality AI models improves the accuracy of failure predictions and demand forecasts. In doing so, it creates greater business value.

However, using and maintaining high quality AI in actual worksites is an extremely difficult task. This is because AI systematization does not involve just the perspectives of system development and operations used in conventional systems, such as ordinary system development, resource status monitoring, and alive monitoring. It also requires perspectives unique to machine learning (ML), in which the quality of AI models can change when the AI's environment changes.

That's why Toshiba has developed the MLOps Platform. This platform, developed through close cooperation between AI model development teams ("AI development teams"), system development teams, and operation teams, rapidly meets customer business needs by effectively and efficiently improving AI systems. 

Ongoing Review Processes Are Essential when Using AI

 When a problem occurs while operating an AI system, it is usually necessary to determine if it is a system-side problem or an AI model problem. This takes a lot of time and effort. Even for AI models which have undergone thorough technological verification by the AI development team, production environments may produce unexpected data. When this happens, the AI model itself needs to be reviewed and revised.

What's more, no matter how high the quality of the AI model that has been developed and deployed, maintaining that quality over the long term is difficult. External factors can, with time, negatively affect the accuracy of AI models. For example, if factors such as air temperatures and equipment aging affect the prediction accuracy of AI models, then AI model relearning will be necessary to reflect these external environmental changes and maintain the accuracy of the model's predictions. In other words, when using AI for business, it is vital to implement an ongoing process of reviewing and revising AI models as necessary.

The system we have developed to continuously implement this process is the MLOps Platform. MLOps stands for "Machine Learning Operations." It refers to all of the processes involved in the management of AI model lifecycles, from development to deployment and operation.

* Deployment: The placement and provision of software (such as AI models) in actual operating environments

The MLOps workflow consists of the following eight processes. Continuously implementing these processes enables AI to evolve in response to changes in the external environment, providing value on an ongoing basis.

  1. Continuous Integration (CI):Testing and implementation preparation for 2 and 3
  2. Learning data acquisition and pre・processing
  3. AI model learning
  4. AI model configuration management
  5. Continuous Delivery (CD):Creating microservices to enable application programming interfaces (API) to call AI models and deploying these microservices in operating environments
  6. AI model operation:Operating the AI model microservices
  7. Usage history/prediction accuracy:Collecting AI model prediction results, etc.
  8. Monitoring:Remotely monitoring AI models

We developed a platform that supports these MLOps and created a workflow that connects all of the processes involved in AI model creation, system development, and system operation. This partially automates the AI model lifecycle, from deployment to operation. The MLOps Platform makes it possible for AI development teams, system development teams, and system operation teams to cooperate smoothly (Fig. 1). 

The MLOps Platform Used by Our "SATLYS" Analytics AI

 Toshiba Digital Solutions’ "SATLYS" analytics AI uses the MLOps Platform. SATLYS is an industrial AI service that brings together the expertise developed by the Toshiba Group through its long manufacturing track record. It covers everything from data analysis and system creation to the managed services involved in operation and maintenance when leveraging AI.

AI service development and operation are performed based on MLOps processes. This makes it possible for the AI itself to evolve based on changes in the external environment and customer business needs. It creates an AI usage framework for continually providing value.

In SATLYS, a monitoring system that uses the MLOps Platform keeps a close eye on the prediction accuracy and usage status of the AI models that are in use. It has a function that alerts the AI operation team if the accuracy of predictions begins to decline.

When the AI operation team receives one of these alerts, it uses the monitoring service to compare the data before and after the drop in prediction accuracy and determines whether AI model relearning will be necessary. If the accuracy needs to be improved, new site data will be used to perform relearning. SATLYS managed services can be used to improve and update the AI models (Fig. 2). 

One of the issues encountered when reviewing and revising AI models is that it is often difficult to produce high quality AI models because large amounts of learning data cannot be acquired. To solve this problem, we've developed learning support technology that makes it possible to develop high quality AI models even with little data. SATLYS provides this technology in the form of a managed service. For example, data that is not labeled as correct is analyzed, and only learning data which would be effective in improving AI model quality is selected. This technology improves the efficiency of labeling correct data. Another SATLYS technology improves efficiency when using a small amount of learning data and a large amount of data that is not labeled as correct.

By combining these AI model monitoring and learning support technologies, SATLYS managed services make it possible to rapidly supply revised AI models that maintain the same quality as was provided before the prediction accuracy degradation occurred. The AI itself can evolve based on changes in the external environment and customer business needs, continually providing value.

We are currently considering using SATLYS managed services for "METALSPECTOR/AI," a solution supplied by Toshiba Digital Solutions. This AI-based metal grade determination service, used in metallographic structure testing, automates the grade determination work performed by experienced inspectors. 

Serving as an AI Hub and Accelerating the Evolution of AI Services

 In the future, the MLOps Platform seeks to serve as a hub for AI. We plan to use the MLOps Platform to offer AI technologies developed by Toshiba's Corporate Research & Development Center for creating high quality AI models.

To achieve this, we are coordinating with Toshiba Group companies involved in projects such as power and social infrastructure projects. We are closely examining which items are important for infrastructure service monitoring and visualization, and developing a variety of AI models.

Through this, we plan to develop AI models with unprecedentedly high levels of quality and to speedily provide customers with infrastructure services that offer significant amounts of added value through the use of AI.

Using the MLOps Platform is making it possible to accelerate the evolution of AI models and provide uninterrupted value.

Toshiba's AI, backed by our advanced technological development strengths, will continue to supply services with even greater added value to customer business, as well as to the infrastructure services and data services of the Toshiba Group.


  • The corporate names, organization names, job titles and other names and titles appearing in this article are those as of February 2021.

>> Related information

Related articles