Please enable JavaScript in your browser.

fltech - 富士通研究所の技術ブログ


Advance Notice for the second blog series about Fujitsu Kozuchi AI core engines

Hello. I am Fukuda from the Fujitsu AI laboratory. Fujitsu provides advanced AI technology through the Fujitsu Kozuchi, and we want to encourage people to try our AI technologies to prove their value and benefit their work. Based on users’ impressions and feedback, we will continue to improve these technologies.

In the last blog series, we introduced 5 technologies in Fujitsu Kozuchi, which we have continued to develop and advance as a result of the inputs we received. This time, I would like to introduce a number of new technologies, as part of a second 5-week blog series. In our Tech Blog features, our intention is to present the technology in an easy-to-understand way, as well as explaining how it works.

I’d like to start by giving you an overview of the engines.

AI core engines introduced in the blog series

1. Fujitsu Auto Data Wrangling: Automatically convert tabular data into an AI-appropriate format using generative AI (To be posted around May 16)

Excel and other tabular data can be represented in many ways, and it is necessary to perform data cleaning and other preprocessing to train AI with the data. This technology automatically estimates the types of tabular data columns (dates, categories, etc.) and converts them into a format AI can handle, reducing tasks by 90% compared with human preprocessing. Additionally, it significantly improves AI accuracy by automatically adding new features and enriching the data.

2. AutoML for Vision: Visual AI Made Easy: Automatically generate AI solutions for visual understanding and explaining (To be posted around May 23)

Creating an AI that can understand and respond to questions about an image requires not only the preparation of large amounts of training data and high-performance computational resources, but also advanced expertise in AI. The technology realizes the highest level of identification accuracy on a small training dataset even when the user has no expertise in AI. When a user inputs a task, the system automatically selects the optimal visual AI model, using large models pre-trained with a variety of data.

3. Composite AI: Solves optimization problems through a chat-like interface (To be posted around May 30)

Generative AI provides answers to various problems in an interactive interface, but a major weakness involves optimization problems such as resource allocation. In addition, solving optimization problems requires experts in the field. This technology automatically formulates optimization problems and delivers the solution by just inputting the optimization problem requirements in natural language, reducing the time needed for optimization prototyping from 1-2 months to just a single day.

4. Smart Enterprise Data Assistant: Business data analysis and visualization using chat mode (To be posted around June 6)

In recent years, we have been increasing the use of various business data such as customer behaviour history and product inventory record. Business data analysis requires advanced analytics or programming skills. This technology allows enterprise business operators to analyse business data in chat mode without any programming or customization. It also adds visualization in the answers to make it easier for users to understand.

5. Traffic Image Surveillance: Automatically detect various status and incidents such as traffic volume and disasters for safety (To be posted around June 13)

It is continuously necessary to monitor various items such as object, event, weather, disaster, camera abnormal in traffic surveillance. This technology provides 27 traffic image analysis functions, including traffic volume, traffic events, road status, weather, and camera abnormal status. Each function delivers a high precision recognition rate (averaging more than 95%) and ensures safety through efficient and low-cost traffic control and management.


With Fujitsu Kozuchi, our objective is to prioritize creating opportunities for trials rather than perfecting the technology itself. We will improve our technology based on your feedback such as "I wish I could do this" or "I want to make this part easier to use." We are publishing our blog series in both Japanese and English, and hope that we can encourage anyone anywhere who is interested in our technology to try out Fujitsu's technology and provide us with your valuable feedback.