TOKYO – Fujitsu today announced a new technology for automated generation of AI solutions, offering users the possibility to customize AI innovation components(1) offered via the “Fujitsu Kozuchi (code name) – Fujitsu AI Platform.” Fujitsu will start offering the new technology via Fujitsu Kozuchi to users in Japan by December 2023 and plans to roll out services to the global market in the future.
Fujitsu’s Kozuchi AI platform, launched in April 2023, already offers users access to a wide range of powerful ready-made AI and ML technologies – leveraging the new technology, users themselves will now be able to further modify and customize AI solutions on the platform using natural language input, without the need for expert knowledge or the support of AI engineers. Applied to optimization problems in production scheduling, the new technology can help to reduce man-hours required for AI model creation by up to 95%.
Based on the newly developed technology, Fujitsu aims to build a system to automatically generate AI technologies for optimization, prediction, and detection of product anomalies, providing customers the optimal combination of innovation components.
The new technology represents part of Fujitsu’s framework for Composite AI for solving customers’ increasingly complex problems by combining different AI innovation components including for functions like demand forecasting and production scheduling. In addition to Fujitsu Kozuchi (code name) – Fujitsu AI Platform, Fujitsu will also offer a framework for external platforms in cooperation with Palantir Technologies Inc. (Palantir). Through these initiatives, Fujitsu aims to contribute to the realization of a sustainable society with AI that can adapt and respond to changes in business and society.
Background
Fujitsu Kozuchi (code name) – Fujitsu AI Platform provides leading-edge AI innovation components and AI core engines, easing the path to applying AI in business operations by enabling faster verification of different potential AI solutions by customers. Working with customers of the new platform, Fujitsu recognized their need to further adjust AI innovation components provided via the platform to their specific business demands. The time-consuming manual adjustment of AI components and modification of prototype technologies by AI engineers has at times contributed to longer lead times in the delivery of completed AI solutions to customers.
To deploy optimal AI solutions more rapidly, Fujitsu developed a new technology that enables customers to modify AI innovation components provided via Fujitsu’s Kozuchi platform by themselves using natural language.
The newly developed technology interprets programs and mathematical expressions converted by Large Language Models (LLMs) and generates the set of solutions that meet customers’ requirements in a graphical format. In this way, the new technology enables the creation of expert-level mathematical expressions. By training AI models with this graph data, the new technology enables the automatic creation of different AI models tailored to customers’ needs, in areas including prediction, optimization, and anomaly detection.
Adding previous learning data to the graph data makes it possible to efficiently re-train AI models even under new conditions (Figure 2). By combining this technology with a LLM, users can rapidly repeat prototyping, modifying, and adjusting of AI solutions through commands in natural language, without requiring advanced AI engineering skills.
About cooperation with Palantir
Fujitsu will leverage Palantir’s Artificial Intelligence Platform (AIP), which allows the use of securely hosted large-language models to help customers solve complex problems for their business. Together, AIP and Fujitsu’s AI Platform and Fujitsu’s framework for Composite AI enables users to find the best models for their data and business problems, automatically build and train those models, and allow users to immediately deploy those models for their business operations and planning.
Figure 2: Features of the newly developed technology
[1] Palantir Technologies Inc. :Headquarters: Colorado, USA; Co-Founder and CEO: Dr. Alexander Karp
