(2021) 🏅 1st place
English / Japanese
I am a Software Engineer with over a decade of work experience in data analysis and R&D across various industries. My interests lie particularly in satellite or aerial imagery, GIS and security applications. I have proven track records in AI competitions hosted by DIU, NGA, and SpaceNet (founded by IQT Labs and Maxar), achieving top placements1,2,3.
Beyond participating in competitions, I am also passionate about designing and managing them. From 2018 to 2023, I served as a committee member for the AI Edge Contest organized by METI and NEDO, contributing to its preparation, operation, and judging processes.
I welcome any inquiries regarding competitions. Feel free to reach out to me at eowner at gmail.com.
Publications
See all my publications on Google Scholar.
Papers (selected)
- Two-stage discriminative re-ranking for large-scale landmark retrieval
Shuhei Yokoo, Kohei Ozaki, Edgar Simo-Serra, Satoshi Iizuka.
CVPR2020 workshop on Image Matching: Local Features and Beyond - Team PFDet's Methods for Open Images Challenge 2019
Yusuke Niitani, Toru Ogawa, Shuji Suzuki, Takuya Akiba, Tommi Kerola, Kohei Ozaki, Shotaro Sano.
ICCV 2019 workshop on Open Images Challenge - Large-scale landmark retrieval and recognition under a noisy and diverse dataset
Kohei Ozaki, Shuhei Yokoo.
CVPR2019 workshop on Landmark Recognition - Using the mutual k-nearest neighbor graphs for semi-supervised classification on natural language data
Kohei Ozaki, Masashi Shimbo, Mamoru Komachi, Yuji Matsumoto.
CoNLL 2011
Books
- The Art of Deep Learning Programming on Kaggle (Japanese)
Kohei Ozaki, Takuya Akiba, Takanori Hayashi, Shotaro Ishihara.
Presentations / Talks
International Conference
- 5th place solution: kNN shortlist and rotation correction
CVPR 2023 workshop on Image Matching: Local Features and Beyond - Local vs Global Descriptor? Relevance Scoring using Both
CVPR 2018 workshop on Large-Scale Landmark Recognition - 3-Stage Ensemble and Feature Engineering for MOOC Dropout Prediction
KDD 2015 workshop (KDD Cup)
Tutorials and Symposium Presentations
- Climbing the Ladder: A Guide to Machine Learning Competitions
ViEW 2022 - Vision Technology Practical Application Workshop, 2022-12-08 - Learning from Winning Solutions in Data Analysis Competitions
Chiba Institute of Technology STAIR Lab Artificial Intelligence Symposium, 2018-03-03 - Data Analysis Competition Techniques and Recent Advancements
Chiba Institute of Technology 14th STAIR Lab Artificial Intelligence Seminar, 2017-10-20 - Python and Practical Applications in Data Analysis Competitions
FIT2016 15th Forum on Information Technology, 2016-09-09 - Techniques and Trends in Competitive Data Mining
Kyoto University Kashima Laboratory Open Seminar, 2015-10-15
University Lectures
- Advanced Information Science
Tokyo Metropolitan University (Intensive Lecture and Exercise), 2019-09-19 - Big Data Analysis
Osaka University (Guest Lecture), 2018-07-24
slides
Projects
Mahjong AI
React
Rust
PyTorch
Google Cloud
Exploring methods for developing strong game AIs through reinforcement learning. My ultimate goal is to leverage the development of strong game AIs for analyzing and discovering game strategies. I am also interested in methods for efficient learning in such complex environments.
Simple self-training often gets stuck in local optima. My research investigates how techniques like feature engineering, importance sampling, curriculum learning, and exploration can enhance learning efficiency.
Other Activities
Data Science Competitions
PyTorch
Python
I have participated in various machine learning competitions, including KDD Cup and Kaggle. I ranked as high as 4th on Kaggle's user rankings and have been awarded a total of 20 gold medals, a recognition given to top performers in Kaggle competitions.
I am always eager to learn new technologies, so I actively participate in competitions held in conjunction with academic conferences and those based on cutting-edge research. I am particularly drawn to competitions involving satellite imagery analysis, which is one of my key areas of interest.
Event Organization (Kaggle Tokyo Meetup)
From 2016 to 2023, I planned and organized Kaggle Tokyo Meetup events. As a community-driven technical exchange, we hosted a total of eight events at various companies' offices. Starting with a small gathering of around six people, the event grew to host over 120 attendees at the Google Tokyo office, thanks to the numerous captivating presentations by our speakers.
You can watch the event on Kaggle Tokyo Meetup 2023 Live Stream (youtube).
Programming Contests
ACM/ICPC is an annual global collegiate programming contest. During my university years, I formed a team with friends from the Tokyo University of Science and participated in the contest, advancing to the Asia Regional Finals twice (2007 Tokyo, 2008 Aizu).
I have also participated in the ICFP Programming Contest organized by ACM SIGPLAN. In 2014, I participated individually under the name "Standard ML/Yeah!" and ranked 6th in the Lightning Round4. That year's challenge was to implement a Pac-Man clone AI running on a LISP machine.
Presentations and Talks
- IMC2023 Recap & Strategies for Code Competitions, 1st Kanto Kaggler Meetup
- AMA with the Author of "Mastering Deep Learning Programming for Kaggle", Weights & Biases Tokyo Meetup #2
- Learning from Data Analysis Competition Solutions, PyCon mini Osaka
- Techniques and Strategies in Image Processing Competitions, Kaggle Student Meetup by Sansan
- Reproducibility and Efficiency in Experiments (Docker and Serialization), Big Data Infrastructure Study Group (University of Tsukuba)
- Workflow, Serialization & Docker for Kaggle, Kaggle Tokyo Meetup #1
- Techniques (Tricks) for Data Mining Competition, BrainPad Internal Presentation
- Data Mining Competition Techniques, PyData.Tokyo Meetup #6
- InfluxDB & LevelDB Inside-out, Monitoring Casual Talk in Kyoto
- Distributed Graph System & Related Topics, 18th Data Mining + WEB Study Group