Hooked on Gaming in His Youth, Now Gold Medal Winner at the International Kaggle AI Competition

Kazuki Kanamoto, who's been with Shimadzu for four years, has won the title of “Kaggle Master”

Winning the gold medal at the “Kaggle” International AI competition on his fifth attempt... meant spending three months thinking intensely about how to make sense of noisy and missing data, with daily trial and error.

We interviewed Kazuki Kanamoto (Imaging Group, Research and Development Department, Analytical & Measuring Instruments Division, Shimadzu Corporation), winner of the “Kaggle Master” title, about this challenge.

 

Eighth Place out of 3,559 Teams! A Solo Challenge for Three Months

Kaggle is the world's largest platform for data scientists, with over 20 million users. The platform is designed such that AI models are proposed in response to tasks presented by companies and researchers, these models compete in terms of accuracy, and the best AI models are exchanged for prizes.

Artificial intelligence (AI) and machine learning (ML) models are mathematical representations constructed using data for the purpose of implementing a specific task. The models learn patterns from the data, and then make predictions and classifications with respect to future data sets.

Schematic diagram of the training sequence for the model created by Kanamoto

The competition that Kanamoto entered is called “Child Mind Institute — Problematic Internet Use.” It was hosted by the Child Mind Institute, and was sponsored by NVIDIA Corporation, Dell Technologies Inc. and other major corporations.

The aim was to predict the severity of internet addiction in adolescents based on a combination of sensor data from wrist-worn devices, questionnaires, and physical measurement scores. It was held for three months from September to December 2024, and 3,559 teams participated.

Although Kanamoto participated on his own, he did amazingly well, finishing in 8th place. He won a gold medal, and was certified as a “Kaggle Master,” which is equivalent to being in the top 1 % of participants in the competition.

Heart & Soul into Repeated Trial and Error

Kanamoto’s reason for taking on this competition challenge was his frustration with his first competition attempt, and his interest in time series data analysis. He noted that the data on this occasion included time series data and noisy data with plenty of missing values, acquired in typical households.

He decided to construct a robust model resilient to noise and missing values, which he improved repeatedly through trial and error. These efforts bore fruit, as his model was highly praised at the competition.

For three months he put his heart and soul into repeated trial and error. In reflecting on his achievements, Kanamoto said that “Running, which I do as strength training, was the ideal time to think through the details of my attempt, and I spent mealtimes as well thinking about ways to increase my score.” When asked about what he attributes his win to, he laughingly replied that “Frankly, it was willpower!”

Power of Concentration and Competitive Spirit Honed Playing Video Games

When asked about where his studies began, Kanamoto said that “I’ve liked video games since I was a kid, and was crazy about various types of them. I liked games so much that it motivated me to go to university to study information science. I entered university thinking that I’d probably be writing games, but there I fell in with computer science, which has led to my current career path.”

“Kaggle has elements that are very similar to games. Just like in online games, the results from competition are reflected in sequential rankings, which is an interesting approach.” Spurred on by the ranking system, which fanned the competitive spirit from his youth hooked on gaming, he has secured success on the world stage.

Benefiting Society with Image Processing Using AI

Kanamoto calls himself an image processing engineer. He is involved in the development of AI-equipped products such as the Multi-data Analysis Platform “PLUS ALGO”, and the Web Application Supporting Cellular Observations “Cell Pocket”.

Here we introduce one example of image processing using AI. PLUS ALGO AM, which uses PLUS ALGO, can determine the quality of soldering on parts mounted on electronic PCB. Analyzing images for the shape of protrusions and the ratio of voids in soldered joints will lead to test automation and error prevention.

In this way, image processing techniques using AI are contributing to society in a variety of ways, at manufacturing sites and in the medical field for example.

Using the X-ray fluoroscopic image of the electronic substrate, chip capacitors on the back are excluded using AI techniques

Determination results

Testing is automated with AI image data analysis, preventing human error and variance in testing quality