- Aiming to realize AI that can share data between different companies while keeping it confidential and maintaining high predictive accuracy -
SB Technology Corp.
SB Technology Corp. (Head office: Shinjuku-ku, Tokyo; President & CEO: Shinichi Ata; hereinafter referred to as SBT) will begin demonstration experiments together with Zeon Corporation (Head office: Chiyoda-ku, Tokyo; President and CEO: Toyoshima Tetsuya; hereinafter referred to as Zeon) with the aim of realizing materials informatics*1 through confidential data sharing between companies using secret computing technology.
This PoC was conducted to verify the application of secure computing technology to an AI model trained on data from two Zeon Group companies*2, after confirming an improvement in the property prediction performance of the AI model.*3 The purpose of this PoC is to realize an AI that allows different companies to share each other's data while keeping it confidential, while maintaining high prediction accuracy.
This will allow the AI to incorporate the knowledge of optimal compounding that was previously relied on experienced researchers, and is expected to further accelerate and streamline research and development. In research and development, component manufacturers have previously been unable to communicate requirements to material suppliers in numerical terms, and have had to repeatedly prototype and evaluate compounds, but by using a confidential property prediction AI system, it is expected that this process will be significantly shortened.
■Background
Since 2021, Zeon has been developing an AI that structures and accumulates experimental data to predict physical properties, but there was a challenge in improving the accuracy of predictions because the amount of training data was insufficient with its own data alone. In addition, although Zeon developed an AI system capable of predicting physical properties that can be used by client companies, the formulation conditions that serve as queries during inference and the physical property prediction results are confidential information to client companies, making it difficult for clients to use Zeon's AI prediction system.
SBT is promoting its business with a focus on cloud, security, AI and data utilization. In promoting co-creation digital transformation with customers, the company has been focusing on the usefulness of secure computation technology, which enables data sharing between different companies, such as in supply chains and within the same industry, and has been conducting basic research since 2022. As part of this, the company has been working on verifying and evaluating software-based secure computation technology and considering business with customers.
This time, SBT discovered the possibility of solving Zeon's problems by using the secure computing technology it had been considering, and decided to undertake a demonstration experiment.
■Outline of demonstration experiment
The anonymized AI system for predicting rubber physical properties consists of two components: a TEE*4 that conceals the AI processing for predicting rubber physical properties, and an environment that separates the web system for each customer.
In this demonstration experiment, the property prediction AI will be kept confidential, and the training data will be only accessible to the provider and will be used for AI training. We aim to create an environment where inference queries and results can only be viewed by inference users via the web. The demonstration experiment is scheduled to run until the end of December 2024.
SBT will be responsible for the overall system architecture design and infrastructure construction, including the TEE, while Zeon will develop applications that require expertise in the rubber industry.
■Future outlook
Going forward, SBT will work with Zeon to aim for commercial use of the confidential property prediction AI system within the Zeon Group.
Through this PoC, SBT will work to use AI technology to share confidential data and help companies solve a wide range of issues that they face when trying to safely use data between different companies.
Contact information regarding this press release
● SB Technology Corp. Public Relations Department
E-mail: sbt-press@tech.softbank.co.jp