Performance

Participation in machine learning contest (Kaggle)
Approximately 118,000 data scientists from all over the world from fields such as information science, statistics, economics and mathematics register and compete for the optimal model of machine learning (AI). Won 4 silver medals and 2 bronze medals.

APTOS 2019 Blindness Detection (Silver Medal Acquisition)
Participated in a contest to detect diabetic retinopathy from fundus camera images. 104th place, top 4% (2,943 team participation) scored and won a silver medal.

[Simple picture to explain Diabetic Retinopathy]

credit : https://www.eyeops.com/

[Usage method]
EfficientNet

Predicting Molecular Properties (Silver Medal Acquisition)
Participated in a contest to predict the interaction between atoms. 44th place, top 2% (2,749 team participation) scored and won the silver medal.

[Usage method]
GCN
LightGBM

Jigsaw Unintended Bias in Toxicity Classification (Silver Medal Acquisition)
Participated in a text classification contest. 105th place, top 4% (3,165 team participation) scored and won the silver medal.

[Usage method]
BERT
LSTM

New York City Taxi Fare Prediction(Can you predict a rider’s taxi fare?)
Participated in a contest to predict New York taxi fare. 18th place, top 2% (1,488 team participation) scored. During the contest period, the world’s top track record was achieved as shown below.

[Usage PC]
Mac Book Pro (CPU:Intel Core i7 2.9GHz, Memory:16GB)


AI (artificial intelligence) related work
Operation failure data AI analysis
The characteristics of individual differences due to differences in the operation of product inspectors, etc. are extracted, the causes leading to defects are analyzed by AI, and the inspection results are evaluated. Based on this, sensor specifications, control specifications, component specifications, etc. will be changed (determined) to realize the development of safer products that do not cause problems.


Software development support for onboard monitoring systems
Ohtsu Maritime Research Institute Co., Ltd. Representative Director Tohru Itoh is developing software for a new navigation / engine hybrid monitoring system using statistical models researched and developed by the late Kohei Ohtsu (Professor Emeritus, Tokyo University of Marine Science and Technology). Technical assistance was provided as a technical advisor to the Ohtsu Maritime Research Institute.

[Related papers]
Jun Wu, Hui Peng, Kohei Ohtsu, Genshiro Kitagawa, Tohru Itoh: ‘Ship’s tracking control based on nonlinear time series model’, Applied Ocean Research 36 (2012), pp.1-11, 2012