Masaya Ogino, a data scientist at SoftBank Corp. (TOKYO: 9434) won a gold medal in the "Feedback Prize - English Language Learning" competition held by Kaggle, the world's largest data science competition platform. The goal of this AI coding competition was to assess the language proficiency of 8th to 12th grade English Language Learners (ELLs) by utilizing a dataset of essays written by them.
AI System Development Department
AI & Data Science Division
AI Strategy Office
As of January 2023, approximately 180,000 people participated in the competition. Competitors who earned a gold medal, along with other qualifying medals, and placed in the top 1%, received the title of “Kaggle Master.” The competitive process encompassed problem-solving, model building and accuracy testing against other participants.
Contestants utilized AI to predict scores on six aspects of English compositions written by ELLs, and were evaluated on the accuracy of their predictions.
Ahead of the competition, Ogino tested various natural language processing models and performed pseudo-labeling tests using external data. His department lets its members allocate 20% of their working hours to skill improvement, and he spent about four to five hours a day preparing for the competition. He was also able to utilize a company GPU server for AI training.
Ogino said he believes he can apply the knowledge and experience gained from the competition to his daily work, especially in situations where training data is not readily available. He also said he is working to stay on top of fast-changing technological developments while applying the latest technology to his work at SoftBank. Ogino’s next goal is to earn the title of "Kaggle Grandmaster" by continuing to participate in competitions.
More recently, Yuki Nakamura, another data scientist that belongs to SoftBank’s AI & Data Science Division, took top honors in the “Predict Student Performance from Game Play” contest held by Kaggle. In July 2023, with three gold medals, he ranked in the top 0.2% of over 200,000 contestants.
The goal of this competition was to predict student performance during game-based learning in real-time. Contestants developed models trained on open datasets of game logs, and specific tasks included predicting whether game players would be able to answer questions correctly in the latter half of the game, based on their in-game behavior history. By innovating to improve accuracy and expedite inference times, Yuki Nakamura secured a gold medal, and his team placed tenth among the 2,051 participating teams.
SoftBank is actively promoting the development of AI systems using table data such as customer analysis and demand forecasting. Based on the knowledge and experience gained from this competition, the AI & Data Science Division will aim to develop even more advanced AI systems.
(Posted on September 13, 2023)
by SoftBank News Editors