Kohei Ozaki


Data Science Competitions & Competitive Programming

I'm naturally attracted to the fields of machine learning and algorithm. I participated in various competitions such as ACM/ICPC, ICFP Programming Contest and many othrer competitions hosted by Kaggle.

Large-Scale World-wide Landmark Retrieval Challenge (2018-2019)

Google-Landmarks is the largest world-wide landmark recognition dataset. This dataset includes over 5 million images of more than 200 thousand different landmarks. Through the competition held in Kaggle, I developed a high-precision data cleaning method that does not require manual annotation costs. My solution was presented at the CVPR'19 landmark recognition workshop.

Challenge of using deep learning with satellite imagery.

I participated in the SpaceNet Building Footprint Detection Challenge (SpaceNet Round2) and won the 1st place prize. I investigated the use of OpenStreetMap for semantic labeling of satellite image and evaluated it on SpaceNet dataset. The detail of my solution is described in my blog post. Also, I provided the first model that CosmiQ Works adds to their geospatial deeplearning pipeline package, solaris.

Kaggle competitions (Since 2010)

kaggle ucsf

I am a Kaggle Grandmaster (an honor to top competitors with outstanding data science skills) with 7 prize winner results. My highest rank is 4th of 530,000+ data scientists on 2016 April.

Kaggle is the largest data science community in the world. Kaggle always provides us great opportunities to learn a wide range of professional skills on machine learning, recommendation, forecasting, natural language processing and image recognition. I played many competitions with talented data scientists across different domains.

KDD Cup (ACM SIG-KDD) 2015

kdd photo

The KDD Cup, sponsored by ACM/SIGKDD, is one of the most prestigious data-mining contest. The task in 2015 was to predict whether a user will drop a course within next 10 days based on his/her prior activities. My team, "InterContinental Ensemble", won the 1st prize.

  • 2015.07: 1st place in ACM SIGKDD, KDD Cup 2015 (out of 821 teams; joint work with Jeong-Yoon Lee, Andreas Töescher, Michael Jahrer, Mert Bay, Peng Yan, Song Chen, Tam T. Nguyen and Xiaocong Zhou.)
  • 2014.07: 7th place in ACM SIGKDD, KDD Cup 2014 (out of 472 teams).

The ICFP Programming Contest 2014


The ICFP Programming Contest is an international programming competition held annually since 1998. The task in 2014 was to write an AI for a simplified pacman game. You could write an AI in any programming language on any platform(s), as long as it compiles to SECD machine instructions for pacman and 8-bit machine assembly instructions for ghosts. I wrote a compiler for pacman in Haskell and finished at 6th place on lightning round (the stage of first 24 hours.)

  • 2014.07: The ICFP Programming Contest 2014 (6th/94 in Lightning Round. My teamname is Standard ML/Yeah!)

Competitive Programming / ACM-ICPC and Google Code Jam


Competitive programming is a mind sport where participants compete to solve a specific problem by coding. ACM/ICPC is an annual multi-tiered competitive programming competition amoung the universities of the world. I joined this contest with two collageate friends at Tokyo University of Science and advanced to the Asia Regional Contest 2 times.

  • 2012.05: Google Code Jam 2012 (Advanced to Round 2)
  • 2008.10: ACM/ICPC Asia Regional Contest 2008, Aizu.
  • 2007.11: ACM/ICPC Asia Regional Contest 2007, Tokyo.