Hitachi data scientist takes second place in global AI data-analysis competition on Kaggle and is promoted to Kaggle Master

July 18, 2023

A team including Wataru Takahara, who is involved in Hitachi, Ltd.'s (hereinafter, Hitachi) Material Development Solution*1, took second place and a gold medal*2 in the Vesuvius Challenge - Ink Detection competition held on the global AI data-analysis competition platform Kaggle. Takahara previously won silver medals in Kaggle's table data, image data, and text data competitions, and with the results of this competition, Takahara was promoted to Kaggle Master*3.

This competition was held from May 16, 2023 to June 15, 2023, Japan time, with 1,249 teams participating. The assigned task was to build an AI model that detects ink from 3D images of ancient scrolls that were carbonized by volcanic eruptions. This AI model would make it possible to restore text even from scrolls that are difficult to decipher due to carbonization. Although it was difficult to build an AI model with high generalization performance due to the small amount of learning data for building AI models, Takahara's team improved generalization performance and achieved highly accurate ink detection by incorporating regularization*4 and data expansion*5 and combining various types of AI models. Even in the field of materials development in which Takahara is involved, there are many cases of building of AI models that require high generalization performance with only a small amount of learning data. The knowledge gained from this competition can be applied in the field of Materials Informatics, which applies AI to materials development, and is expected to lead to strengthening of the technical capabilities of Material Development Solution.

The Public Platform Solution Operation Public Digital Transformation Department of the Government & Public Corporation Information Systems Division promotes participation in competitions such as Kaggle to improve data analysis technology, and data scientists including Takahara are working to improve data analysis technology. Hitachi will continue to promote data analysis initiatives and contribute to the improvement of social, environmental, and economic value.

*1:

 A service that uses technologies such as Materials Informatics to support development of new materials for customers.

*2:

 The number of teams that can win gold medals changes according to the number of teams participating in the competition. If there are less than 250 teams, gold medals will be awarded to the top 10 teams, and if there are 250 teams or more, gold medals will be awarded to the top 10 teams + 0.2%. In this competition, gold medals were awarded to the top 12 teams.
Reference: https://www.kaggle.com/progression

*3:

 The rank is determined according to the number of medals obtained. To become a Kaggle Master, you need to obtain 1 gold medal and 2 silver medals. As of July 2023, Kaggle has more than 14 million registered users worldwide, and Kaggle Master is a difficult title to reach, with only about 2,000 of those registered being Kaggle Masters.

*4:

 A method to prevent AI models from excessively adapting to learning data.

*5:

 A method to expand the amount of learning data by inverting image data for learning, etc.

Wataru Takahara, runner-up in the global AI data-analysis competition on Kaggle

Related information

Contact information

Hitachi Ltd.
DIGITAL SYSTEMS & SERVICES SALES MANAGEMENT DIVISION, Government & Public Corporation Information Systems Sales Management Division, Public Business Planning Operation
Contact person : Takemasa