Curriculum Vitae

A Ph.D. student in Informatics with experience in mathematical optimization in academia and machine learning and web development in the industry. Obtained a Bachelor of Science (B.Sc.) in Computer Science (2018), Master of Science (M.Sc.) in Computer Science (2019). Also expected to get a Doctor of Philosophy (Ph.D.) in Coumter Science (2022).

Education

Nomenclature Note: In Japan, Doctorate Degree = Doctor of Philosophy (Ph.D.) = Doctor of Science (D.Sc.), Master’s Degree = Master of Science (M.Sc.), Bachelor’s Degree = Bachelor of Science (B.Sc.), and Informatics = Information Science = Computer Science are often used interchangeably. For this reason, Ph.D., M.Sc., B.Sc., and Computer Science are used here as universal translations to reduce confusion. Blame someone in the Middle Ages for this mess.

Oct 2019 – Mar 2022 (Expected) | Ph.D. in Computer Science

Kyoto University (Japan)

GPA: 4.1

Apr 2018 – Sep 2019 | M.Sc. in Computer Science

Kyoto University (Japan)

GPA: 3.768

Chosen as the valedictorian for the degree conferment ceremony.

Thesis: Merit functions for multiobjective optimization and convergence rates analysis of multiobjective proximal gradient methods.

Abstract: We first presented two new merit functions for nonlinear multiobjective optimization, which extend the one defined for linear multiobjective optimization. These functions return zero at the solutions of the original problem and strictly positive values otherwise. Furthermore, by examining these merit functions’ properties, We showed sufficient conditions for the existence of weakly Pareto optimal solutions and Pareto optimal sets’ boundedness. Finally, we analyzed the convergence rates of the recently proposed multiobjective proximal gradient methods by using these functions. We showed that both methods with and without line searches have a sublinear convergence rate for non-convex and convex cases. We also proved that the algorithm without line searches converges linearly in the strongly convex case.

Apr 2014 – Mar 2018 | B.Sc. in Computer Science

Kyoto University (Japan)

GPA: 3.836

Chosen as the valedictorian for the graduation ceremony.

Thesis: Proximal gradient methods for multiobjective optimization and their application to robust multiobjective optimization.

We proposed new descent methods for unconstrained multiobjective optimization problems, where each objective function is the sum of a continuously differentiable function and a closed, proper, and convex but not necessarily differentiable one. The methods extend the well-known proximal gradient algorithms for scalar-valued optimization, which are efficient for particular problems. Moreover, we proved that each accumulation point of the sequence generated by these algorithms, if it exists, is Pareto stationary. We also presented their application in robust multiobjective optimization, which is a problem that considers uncertainties. In particular, we showed the way of converting the subproblems of the proposed algorithms into quadratic, second-order conic, or semidefinite programming problems. Finally, we also carried out some numerical experiments with Matlab.

Experience

Apr 2020 – Present | Japan Society for the Promotion of Science

Research Fellowship for Young Scientists (DC1)

Awarded to excellent Ph.D. students, this fellowship offers the fellows an opportunity to focus on a freely chosen research topic based on their own innovative ideas. This program’s recruitment phase has a very narrow selection process with a selection ratio of 20%.

Research subject: Develop efficient and practically applicable algorithms for non-convex and convex multiobjective optimization problems. Analyze convergence and convergence rate of them through theoretical proof and numerical experiments via MatLab.

Feb 2017 – Present | Self-employed

YouTuber

Manage two channels, the Let’s Play channel with 18,000 subscribers and the Music Cover channel with 5,000 subscribers. One of the videos in the Let’s Play channel has more than 1 million views.

Sep 2020 – Sep 2020 | VOYAGE GROUP Inc.

Software Engineering Internship

Developed a coding test tool in a four-person engineering team to improve efficiency in HR. Communicated with the HR staff and asked about their recruiting problems. Designed the prototype with Miro. Implemented authentication by Google account, uploading source codes, and grading each submission, using Vue.js and Firebase.

Aug 2020 – Sep 2020 | Yahoo Japan Corporation

Data Scientist Internship

Estimated heterogeneous treatment effect on earnings of Yahoo! Japan Auction of a recommendation system by using Machine Learning with Python on Jupyter Notebook. Analyzed seven days of preprocessed traffic data of Yahoo! Japan Auction, including the click data of the recommendation banner, in the form of Pandas DataFrame. Implemented X-Learner, a kind of algorithms called “meta-learner,” which estimates CATE (Conditional Average Treatment Effect) using any machine learning estimator. Used LightGBM and linear regression as the estimator of X-Learner. Prevented leakage through correlation analysis. Verified if the recommendation system impacts the earnings via Student’s t-test and Cohen’s d effect size. Note that estimating the treatment effect is strongly related to so-called “A/B testing.”

Aug 2020 – Aug 2020 | Cookpad Japan Inc.

Software Engineering Internship

Built a new Ruby on Rails web application from scratch. It enables users to record and rate recipes they have used. Using the Lean Startup Methodology, developed the MVP (Minimum Viable Product) that satisfies the value hypothesis and utilized a feedback loop through user testing. Presented a prototype of the product by using Figma. Implemented functions of recording recipes with title, memo, image, and rating, editing and destroying them, showing them in a grid view, and sorting them by rating and date, using Ruby on Rails, Docker, Boostrap.

Feb 2019 – Mar 2020 | HACARUS Inc.

Designed and developed some machine learning models for the manufacturing and agricultural industry, mainly through analyzing a small number of image datasets, typically using Python. In particular, developed models for surface defect detection through various techniques, including dictionary learning and fused lasso and hyperspectral image segmentation via online dictionary learning and typical classification algorithms (e.g., support vector machine, random forest). Surveyed scientific papers and implemented algorithms from them, using some machine learning libraries (e.g., scikit-learn), image processing libraries (e.g., open-cv, scikit-image), data visualization libraries (e.g., matplotlib, seaborn), and notebooks environments (e.g., Jupyter).

Other minor engagements included: Added some new algorithms (e.g., alternating direction methods of multipliers for the generalized lasso, matching pursuit for sparse coding) to the open-source library “spm-image,” a scikit-learn compatible library of sparse modeling and compressive sensing. Implemented batch-OMP, an improved version of orthogonal matching pursuit algorithm, to contribute to the research project from NEDO (the New Energy and Development Organization), the Japanese largest governmental R\&D organization. Developed front-end of an in-house web application that generates an interactive visualization of image datasets’ quality, using flask, pandas, and bokeh. Wrote an overview article of pliable lasso and hmlasso, state-of-the-art algorithms for sparse modeling, for Codezine, a Japanese web magazine for software developers.

Languages

  • English: Professional working proficiency
  • Japanese: Native or bilingual proficiency

Publications

タイトルとURLをコピーしました