Skip to content

Latest commit

 

History

History
493 lines (400 loc) · 21.8 KB

resume.md

File metadata and controls

493 lines (400 loc) · 21.8 KB
Raw

MASAHIRO SAKAI (酒井 政裕) ― Research and Development Engineer

Masahiro Sakai is a researcher in the field of theoretical computer science. He is working as a software engineer at Preferred Networks, Inc. He received a master's degree in “Media and Governance” with “Cyber Informatics” concentration at Keio university. He translated “Software Abstractions” and “Types and Programming Languages” into Japanese with his colleagues.

He is passionate in learning new technologies / skills / knowledge and putting them into practice by writing code, and he strives hard toward perfection.

His interest includes Programming Language Theory, Functional Programming, Software Engineering, Category Theory, Constraint Solving, SAT/SMT Solvers, Machine Learning, Mathematical Optimization, and Computer Security.

Professional Experience

Preferred Networks, Inc.

Engineer (May 2017 to Present), Engineering Manager (December 2019 to May 2022), Tech Lead (June 2022 to February 2023), and Scrum Master (May 2023 to Present):

I worked as a software engineer applying cutting-edge deep learning techniques to solve highly challenging real-world problems.

  • Automated operation of petrochemical plants (2019-):

    Petroleum refineries and petrochemical plants are large and complex, and their control requires operational know-how based on many years of experience. We first achieved the automated operation of a butadiene extraction plant and later the automated operation of a crude distillation unit, which is more difficult and has a greater economic impact. Our system is used in ENEOS's Kawasaki Refinery 24/7. https://www.preferred.jp/ja/news/pr20211202/ https://www.preferred.jp/ja/news/pr20230731/ https://www.preferred.jp/ja/news/pr20240524/

    I made many important contributions early in the project, which are still in use: an overall control scheme, inductive bias for integrating domain knowledge into neural network models, and communication with DCS (Distributed Control System). I was involved in many other parts of the project, including simulator-related R&D and system development (front and backend). I also supported the project as an engineering manager and a scrum master.

  • Several PoC projects for using machine learning in industrial domains (2017-2019):

    I worked on multiple anomaly detection projects in the factory (e.g. anomaly detection of cutting tools). I adapted already developed algorithms (neural network models) to the customer's environment and solved problems including specialized data preprocessing and data augmentation, fixing memory leak issues in deployed environments, etc. Our technology is incorporated in FANUC's products as “AI Servo Monitor” (AIサーボモニタ). https://www.fanucamerica.com/products/cnc/cnc-software/machine-tool-data-collection-software/ai-servo-monitor

  • Menoh DNN inference library (2018-2019):

    Menoh is a DNN inference library that loads trained DNN models from the ONNX format and runs inference at high speed on an Intel CPU using MKL-DNN (oneDNN). The main developer is Shintarou Okada -san and I contributed with code reviews, CI, windows/macOS support, Haskell binding, etc. I also proposed the name “Menoh” (瑪瑙). https://tech.preferred.jp/ja/blog/menoh-release/

  • Research for applying deep learning technologies in the field of automated theorem proving (ATP) (2017-2019):

    We tried to apply deep neural networks to theorem-proving related problems including premise selection in automated theorem proving of first-order logic, inference rule selection in intuitionistic propositional logic, NN-based heuristics in SAT solver, etc. https://www.slideshare.net/slideshow/proof-summit-2017-for-slideshare/78254931 https://github.com/pfnet-research/chainer-formulanet https://arxiv.org/abs/1811.00796

Toshiba Corporation

Engineer (Apr 2007 to June 2015) and Research Scientist (July 2015 to April 2017):

I worked as a researcher for the System Engineering Laboratory, Toshiba Research and Development Center. I have worked in software dependability team to develop technology based on formal methods, program analysis, and automatic test-case generation to achieve more dependable systems. I have also worked in cities infrastructure solutions team and developed a recommendation system that recommends brick-and-mortar shops/goods on smartphones for regional promotion.

  • Economic load dispatch (2015-2016):

    Economic load dispatch is the determination of the output of electricity generation units that meet the electricity demands with the lowest possible cost, subject to complex operational constraints. I assisted research project of economic load dispatch problem by removing performance bottlenecks of our prototype optimization system, setting up continuous integration environment, and writing heuristic algorithm and minor modules of the system. My work is partly applied to the system for TEPCO Fuel & Power, Incorporated. (c.f. https://www.global.toshiba/jp/news/energy/2016/10/news-20161026-01.html https://www.global.toshiba/content/dam/toshiba/migration/corp/techReviewAssets/tech/review/2016/03/71_03pdf/0A.pdf#page=6 (written in Japanese))

  • Recommendation system for the Kawasaki Grand City Mall project (2013-2015):

    Toshiba carried out a demonstration experiment on smart city and O2O service around Kawasaki Station (c.f. https://www.global.toshiba/jp/news/corporate/2013/12/pr1901.html http://www.city.kawasaki.jp/200/page/0000053904.html (written in Japanese)). As part of this experiment, we have developed a recommendation system that recommends brick-and-mortar shops/goods on smartphones for regional promotion. I led a team for data analysis and recommendation algorithm development.

  • SMT-based semi-automatic test-case generation for power plant control systems (2010-2014):

    Power plant controllers are often programmed using graphical data flow language such as Function Block Diagram (FBD). We have proposed new test coverage criteria for FBD and developed technology to automatically generate test cases that satisfy the proposed criteria. I developed core algorithms of our approach that encodes both FBD programs and conditions for proposed criteria into logical formulas and solves the conjoined formula using SMT solvers to generate test cases.

  • Molatomium (2009-2011):

    I assisted language/runtime design of “Molatomium”, a parallel programming model for “Cell Broadband Engine”-like heterogeneous multicore processors, which was shipped with Toshiba's flagship digital TV “CELL REGZA”. (c.f. https://www.usenix.org/conference/hotpar-10/molatomium-parallel-programming-model-practice )

  • Specification mining based on UNSAT core enumeration (2008-):

    Specification mining tackles software maintenance and reliability issues by mining latent specifications from source code. We have developed a novel technology to mine pre-condition (one type of specification) by combining model checking and minimal unsatisfiable core enumeration.

  • CForge: Bounded model checker for C language (2008-2012):

    CForge is a C program analyzer based on a program analysis framework Forge developed by the Software Design Group (SDG) at MIT, and we developed CForge as part of a joint research project with MIT SDG. I worked as a main developer of CForge, and worked on its overall design, design of our formal specification language for C, implementation of a kind of C compiler, and so on.

  • Verification of workflow networks based on model checking (2007).

Skills

  • Engineering Knowledge and Interests:
    • Functional Programming, Language Design, Formal Methods, Model Checking, SAT/SMT solvers, Machine Learning, Recommendation System, Mathematical Optimization, Category Theory, Type Theory, Logic.
  • Programming Languages:
    • Haskell, Ruby, Python, Java, C.
  • Languages:
    • Japanese: Native Speaker
    • English: Professional Proficiency
    • German: Basic level (CEFR A2)

Qualifications

  • Coursera

    • Application Development with Cloud Run (Jan. 2023) by Google Cloud Training
    • Securing and Integrating Components of your Application (Jan. 2023) by Google Cloud Training
    • Getting Started With Application Development (Jan. 2023) by Google Cloud Training
    • Getting Started with Google Kubernetes Engine (Dec. 2022) by Google Cloud Training
    • App Deployment, Debugging, and Performance (Dec. 2022) by Google Cloud Training
    • Google Cloud Fundamentals: Core Infrastructure (Dec. 2022) by Google Cloud Training
    • Oil & Gas Industry Operations and Markets (Apr. 2020) by Prof. Lincoln Pratson, Duke University
    • Process Mining: Data science in Action (May. 2015) by Prof. Wil van der Aalst, Eindhoven University of Technology
    • Discrete Optimization (May. 2015) by Prof. Pascal Van Hentenryck, The University of Melbourne
    • Gamification (Mar. 2015) by Assoc. Prof. Kevin Werbach, University of Pennsylvania
    • Heterogeneous Parallel Programming (Mar. 2014) by Prof. Wen-mei W. Hwu, University of Illinois at Urbana-Champaign
    • Introduction to Recommender Systems (Dec. 2013) by Prof. Joseph A. Konstan and Assist. Prof. Michael D. Ekstrand, University of Minnesota
    • Maps and the Geospatial Revolution (Aug. 2013) by Assist. Prof. Anthony C. Robinson, The Pennsylvania State University
    • Developing Innovative Ideas for New Companies: The First Step in Entrepreneurship (May. 2013) by Dr. James V. Green, University of Maryland, College Park
    • Model Thinking (Aug. 2013) by Prof. Scott E. Page, University of Michigan
    • Machine Learning (Jul. 2013) by Assoc. Prof. Andrew Ng, Stanford University
    • Linear and Discrete Optimization (Apr. 2013) by Prof. Friedrich Eisenbrand, École Polytechnique Fédérale de Lausanne
    • Data Analysis (Mar. 2013) by Assist. Prof. Jeff Leek, Johns Hopkins University
    • Computing for Data Analysis (Jan. 2013) by Assoc. Prof. Roger D. Peng, Johns Hopkins University

  • TOEIC Score 960 (Listening: 490; Reading: 470) (Mar. 2018)

  • ドイツ語技能検定試験 3級 (German Diploma in Japan, Basic level) (Feb. 2024)

  • C言語ベースの組込みハードウェア設計 (A course on C-based design of embedded hardware) (Oct. 2007) by NEXCESS (Nagoya University Extension Courses for Embedded Software Specialists).

  • Google Cloud Certified Professional Cloud Developer (Feb. 2023) by Google Cloud

Education

  • March 2007: Master of Media and Governance (Course: Cyber Informatics), Graduate School of Media and Governance, Keio University.

    • Master thesis: "Fusion Transformations Applicable to Non-Strict Functions" ("非正格関数に対して適用可能な融合変換"). Adviser: Prof. Tatsuya Hagino.

  • March 2005: Bachelor of Arts in Policy Management, Faculty of Policy Management, Keio University.

Books and Selected Publications

  • Benjamin C. Pierce, Eijiro Sumii, Yusuke Endoh, Masahiro Sakai, Keigo Imai, Yusuke Kuroki, Yoshihiro Imai, Takafumi Saikawa and Takeo Imai, “型システム入門－プログラミング言語と型の理論－” (Japanese translation of “Types and Programming Languages”). Ohmsha, 2013. ISBN: 978-4-274-06911-6

  • Daniel Jackson, Shin Nakajima, Takeo Imai, Masahiro Sakai, Yusuke Endoh and Yoshio Kataoka, “抽象によるソフトウェア設計－Alloyではじめる形式手法” (Japanese translation of “Software Abstractions: Logic, Language, and Analysis”). Ohmsha, 2011. ISBN: 978-4-274-06858-4

  • Momoko Hattori, Shimpei Sawada, Shinichiro Hamaji, Masahiro Sakai, Shunsuke Shimizu, "Semi-static type, shape, and symbolic shape inference for dynamic computation," In MAPL 2020: Proceedings of the 4th ACM SIGPLAN International Workshop on Machine Learning and Programming Languages, Jun. 2020, pp. 11–19, https://dl.acm.org/doi/abs/10.1145/3394450.3397465

  • Mitsuru Kusumoto, Keisuke Yahata, and Masahiro Sakai, "Automated theorem proving in intuitionistic propositional logic by deep reinforcement learning," Nov. 2018. https://arxiv.org/abs/1811.00796

  • Masahiro Sakai and Takeo Imai, “Interplay between SAT and other Constraint Problems,” Computer Software, Vol. 32, No. 1, 2015, pp. 103-119. doi:10.11309/jssst.32.1_103 https://www.jstage.jst.go.jp/article/jssst/32/1/32_1_103/_article/-char/en/

  • Kohei Maruchi, Hiromasa Shin and Masahiro Sakai, "MC/DC-like Structural Coverage Criteria for Function Block Diagrams," In proceedings of Software Testing, Verification and Validation Workshops (ICSTW). Cleveland, Ohio USA, 2014. pp. 253-259. doi:10.1109/ICSTW.2014.27 http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6825670

  • Takeo Imai, Masahiro Sakai, Masami Hagiya, “An Inference Method of Quasi-Weakest Preconditions by Minimal Unsatisfiable Core Enumeration,” Computer Software, Vol. 30, No. 2, 2013, pp. 207-226. doi:10.11309/jssst.30.2_207 https://www.jstage.jst.go.jp/article/jssst/30/2/30_2_207/_article/-char/en/

  • Masahiro Sakai, Kohei Maruchi and Takeo Imai, "Model-checking C programs against JML-like specification language," in Proceedings of the 19th Asia-Pacific Software Engineering Conference (APSEC2012), Hong Kong, 2012, pp. 174-183. doi:10.1109/APSEC.2012.68 http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6462652

  • Masahiro Sakai, "Fusion Transformations Applicable to Non-Strict Functions," ("非正格関数に対して適用可能な融合変換,") Master thesis, Graduate School of Media and Governance, Keio University, Jan. 2007.

Recognition and Awards

  • 2017 (社内表彰): 株式会社東芝 研究開発センター 業績賞優秀賞 (平成29年4月6日)
  • 2015: the 19th Research Paper Award from Japan Society for Software Science and Technology for the paper “An Inference Method of Quasi-Weakest Preconditions by Minimal Unsatisfiable Core Enumeration” (日本ソフトウェア科学会第19回研究論文賞)
  • 2014 (社内表彰): 株式会社東芝 コミュニティ・ソリューション社 知的財産活動奨励賞 (平成26年7月7日)

Professional Services

  • 日本ソフトウェア科学会 機械学習工学研究会（MLSE） 発起人

  • 情報処理学会 学会誌「情報処理」 編集委員 (2021 to present)

  • 第21回プログラミングおよびプログラミング言語ワークショップ (PPL 2019) プログラム委員

  • External reviewer of IPSJ/SIGSE Software Engineering Symposiums (SES2014, SES2015).

  • Reviewed some technical books (about computer science and functional programming) before publication:

    • Graham Hutton and Kazuhiko Yamamoto, “プログラミングHaskell” (Japanese translation of “Programming in Haskell”), Ohmsha, 2009. ISBN: 978-4274067815

    • Jeremy Gibbons, Oege de Moor, and Nobuo Yamashita, “関数プログラミングの楽しみ”(Japanese translation of “the fun of programming”), Ohmsha, 2010. ISBN: 978-4274068058

    • Richard Bird and Nobuo Yamashita, “関数プログラミング入門 ― Haskellで学ぶ原理と技法―” (Japanese translation of “Introduction to Functional Programming with Haskell), Ohmsha, 2012. IBSN: 978-4274068966

    • Tom Stuart, Koichi Saasda and Takashi Sasai, “アンダースタン ディング コンピュテーション--単純な機械から不可能なプログラムまで” (Japanese translation of “Understanding Computation -- From Simple Machines to Impossible Programs”), O'Reilly Japan, 2014. ISBN: 978-4873116976

    • Richard Bird and Nobuo Yamashita, “関数プログラミング 珠玉の アルゴリズムデザイン” (Japanese translation of “Pearls of Functional Algorithm Design”), Ohmsha, 2014. ISBN: 978-4274050640

    • Mark C. Chu-Carroll and cocoatomo, “グッド・マス ギークのための数・論理・計算機科学” (Japanese translation of “Good Math: A Geek's Guide to the Beauty of Numbers, Logic, and Computation”), Ohmsha, 2016. ISBN: 978-4-274-21896-5

    • Tomoshi Otsuki and Youichiro Miyake, “最強囲碁AI アルファ碁 解体新書 増補改訂版 アルファ碁ゼロ対応 深層学習、モンテカルロ木探索、強化学習から見たその仕組み” (A commentary on AlphaGo and AlphaGo Zero), Shoeisha, 2018. ISBN: 9784798157771

Open Source Activities

  • toysolver / toysat

    Solver implementation of various problems including SAT, Max-SAT, PBS (Pseudo Boolean Satisfaction), PBO (Pseudo Boolean Optimization), MILP (Mixed Integer Linear Programming) and non-linear real arithmetic. In particular, it contains a moderately-fast pure-Haskell SAT solver 'toysat'. I submitted it to solver competitions (Pseudo Boolean Competition and Max-SAT Evaluations) and ranked high in some competition categories.

  • Ruby-GNOME2

    Ruby-GNOME2 is a set of Ruby language bindings for the GNOME >=2.0 development environment. I was one of core developers in the early days. In particular, I have designed/wrote a general bridge between Ruby's object system and GObject object system using reflection/introspection features and replaced non-extensible hard-coded ones in its predecessor Ruby/GTK1.

  • PTQ : An implementation of Montague's PTQ in Haskell

    Montague grammar and the PTQ (The Proper Treatment of Quantification in Ordinary English) by Richard Montague were pioneering work in the field of formal semantics of natural languages, and showed that natural languages (like English) and formal languages (like programming languages) can be treated in a similar way. I implemented his theory in an interactive environment.

  • CPL : A categorical programming language interpreter

    I implemented an interpreter of CPL, a functional programming language based on category theory, designed by Prof. Tatsuya Hagino. In CPL, data types are declared in a categorical manner by adjunctions. Each data type is declared with its basic operations or morphisms. Programs consist of these morphisms, and the execution of programs is the reduction of elements (i.e. special morphisms) to their canonical form.

Other Activities

  • Google Code Jam Japan 2011: 44th place among 918 participants

  • Computer security competitions (CTF; Capture the Flag)

    • CTF for beginners 2015 Tokyo: 6th place among 103 participants
    • Trend Micro CTF Asia Pacific & Japan 2015 Online Qualifier: 46th place among 881 teams, as a team "Ten*48"
    • SECCON 2015 Online CTF: 43th place among 1251 teams, as a team "Ten*48"
    • DEF CON CTF Qualifier 2016: 164th place among 1335 teams, as a team "Ten*48"
    • HITCON CTF 2016 Quals: 34th place among 1024 teams, as a team "Ten*48"

  • Study group of category theory (圏論勉強会): 2004 - 2012

    Category theory is a highly general and abstract mathematical theory, which has applications not only in mathematics but also in theoretical computer science, functional programming, physics, linguistics, and many other fields. I was one of core members of the study group and we studied following materials:

    • “Conceptual Mathematics: A First Introduction to Categories” by F. William Lawvere and Stephen Hoel Schanuel,
    • “Categories, Types, and Structures: An Introduction to Category Theory for the Working Computer Scientist” by Andrea Asperti and Giuseppe Longo,
    • “The Haskell Programmer's Guide to the IO Monad — Don't Panic” by Stefan Klinger,
    • “Temperley-Lieb Algebra: From Knot Theory to Logic and Computation via Quantum Mechanics” by Samson Abramsky
    • “A survey of graphical languages for monoidal categories” by Peter Selinger
    • “Categorical Logic and Type Theory” by Bart Jacobs.

    There is a Japanese article about this study group: “Technical Study Session Map in Japan - Join a Live Session of Engineers!”, 情報処理, Vol.52, No.4, Apr. 2011. http://id.nii.ac.jp/1001/00073853/

  • RHG reading group (RHG読書会): 2003 - 2009

    This study group was a group of programming language geeks. It was started initially as a study group for reading “Rubyソースコード完全解説” (known as “Ruby Hacking Guide”), but eventually, we studied many programming and programming-language related books/materials:

    • YARV (Ruby >=1.9 VM),
    • なでしこ (a Japanese programming language),
    • MinCaml,
    • “ふつうのLinuxプログラミング--Linuxの仕組みから学べるgccプログラミングの王道”,
    • “ふつうのHaskellプログラミング--ふつうのプログラマのための関数型言語入門”,
    • “入門JavaScript”,
    • “実践Common Lisp” (Japanese translation of “Practical Common Lisp”),
    • “ふつうのコンパイラをつくろう--言語処理系をつくりながら学ぶコンパイルと実行環境の仕組み”.

  • Teaching Assistant

    • 2014-2015: Toshiba Systems & Software Institute
    • Feb. 2006: ベトナム高度IT教育者実践研修
    • Aug. 2003 - Sep. 2003: ITスキル標準人材育成研修