#Japan #Generative AI #Tokyo #EMUNI #Sumitomo Electric

EMUNI and Sumitomo Electric: A Partnership in Generative AI Projects

In a recent announcement, EMUNI, a startup originating from Kyoto University’s Matsuo Lab, revealed the successful completion of two significant projects in collaboration with Sumitomo Electric Industries, a leading player in the electrical industry. These initiatives aim to leverage generative AI in the realm of research and development, specifically focusing on the automation of data extraction from academic papers and the enhancement of electronic laboratory notebook inputs.

1. Introduction to the Collaborative Projects

EMUNI, headquartered in Bunkyo, Tokyo, has teamed up with Sumitomo Electric, located in Osaka, to tackle challenges faced by researchers in materials science. The projects, titled "Automatic Extraction Tool for Material Properties from Academic Papers" and "Proof of Concept for Automatic Input into Electronic Laboratory Notebooks," aim to reduce the time and effort researchers spend on data aggregation and input tasks.

2. Development of the Automatic Extraction Tool

2.1 Background

In the field of materials science, researchers frequently review vast amounts of academic literature. This task becomes tedious and time-consuming, especially when extracting essential data such as experimental conditions and material properties, which are often formatted inconsistently across different papers. Traditional keyword search methods have proven inadequate, leading to the need for innovative solutions.

2.2 Approach

The development of the extraction tool aimed to achieve high versatility and precision without imposing strict limits on the types of material properties that could be extracted. To facilitate the customization of the extraction process, user-defined groups were established, and initial definitions and extraction cues were provided using LLM (Large Language Model) technology. Users could then refine these definitions, significantly streamlining the setup process. The tool generates extraction results in both CSV and PDF formats, linking the extracted results with the corresponding sections in the source documents for easy verification.

2.3 Results

A user-friendly interface was created, allowing users to configure groups, upload PDFs, download results, and verify extraction outputs seamlessly. The system demonstrated solid performance, accurately extracting material properties across a variety of categories while managing to address specific user needs effectively.

Points of Innovation

1. High Versatility: By allowing users to define properties independently, the tool could extract a wide array of material characteristics with minimal input burden.
2. Ease of Reference Checking: The LLM's reasoning ability enabled users to efficiently navigate and confirm the basis for the extraction results, enhancing reliability.

3. Proof of Concept for Automatic Input into Electronic Laboratory Notebooks

3.1 Background

As industries like pharmaceuticals increasingly adopt electronic laboratory notebooks for data collection, challenges arise, particularly within manufacturing sectors. The complexity and variability of experimental protocols often lead to increased input burdens, resulting in low data collection rates. The challenge is not just to input data but to enable seamless integration that meets user expectations.

3.2 Approach

In addressing this, EMUNI utilized LLM technology to extract and standardize information from user-defined Excel files, automatically populating the specified formats within electronic lab notebooks. This approach went beyond simple data extraction, as it also involved processing intricate calculations specified by users, yielding refined and actionable data that was stored in a database for future searches and analyses.

3.3 Results

The extraction accuracy from Excel was a remarkable 100% for standard data entries. For more complex calculations tested across 14 use cases, accuracy ranged from 80% to 100%. Furthermore, verifying different formats yielded a 75% to 90% accuracy for complex calculations, showcasing the tool's capability to generalize across varying input conditions.

Points of Innovation

1. Excel Data Handling: EMUNI enhanced the interpretability of Excel data, allowing LLMs to understand its two-dimensional structure effectively.
2. Improving Precision: To enhance the system's handling of complex computations, multiple stages of output were implemented, leading to substantial accuracy improvements.
3. Accelerating Verification: A specialized application was developed to streamline the verification cycle, enabling rapid PDCA (Plan-Do-Check-Act) iterations.

4. Comments on the Collaboration

4.1 Mr. Tatsuya Takakuwa, Sumitomo Electric

Mr. Takakuwa emphasized the critical importance of mastering rapidly evolving LLM technologies in the manufacturing sector. Sumitomo Electric has actively integrated materials informatics to streamline material development. The partnership with EMUNI, characterized by innovative thinking and swift implementation, marks a significant step towards defining new development styles in manufacturing.

4.2 Mr. Yuta Shimono, EMUNI

Mr. Shimono expressed great pride in collaborating with Sumitomo Electric to address inherent operational challenges using cutting-edge AI technology. As a startup from Kyoto University, he aims to establish a model for manufacturing firms in the Kansai region, highlighting the potential for impactful collaboration with Sumitomo Electric in the future.

5. About Sumitomo Electric Industries

• - Core Business: The development of electrical cables and wires with contributions across various sectors, including environmental energy, information and communications, automotive, electronics, and industrial materials.
• - Headquarters: 4-5-33 Kitahama, Chuo-ku, Osaka
• - Website: Sumitomo Electric

6. About EMUNI

EMUNI operates under the mission of "Creating a Happier Work Environment through AI, Sparking Excitement across the World." The company specializes in developing and implementing customized AI solutions tailored to the needs of manufacturing industries.

• - CEO: Yuta Shimono
• - Headquarters: 6-25-14 Hongo, Bunkyo, Tokyo
• - Website: EMUNI
• - Contact: [email protected]

As the collaboration between EMUNI and Sumitomo Electric grows, it is set to redefine the future of research processes in materials science and beyond.