Resume of Hiroyoshi Komatsu

Education

Programming Languages (proficient): Python, Scala, Java.
Programming Languages (barely readable): C/C++, Haskell, JavaScript.
Others: Android apps development, SQL, NoSQL, SPARQL.
7-year experiences and understanding of natural language processing with both deep learning approaches & traditional statistical machine learning models.
Understanding and basic research experience of mathematical logic (proof theory), computational complexity theory, and theoretical cryptography.
Basic skills: Linux, git, kubernetes

A research assistant. I’ve worked on designing and making a dataset for contextual semantic parsing task based on real search queries and making a deep learning system with PyTorch.

Project Details: https://www.ipa.go.jp/jinzai/mitou/2013/gaiyou_s-4
I’ve proposed a new semantic parsing approach for virtual assistant applications. I’ve Implemented a core machine learning algorithm (log-linear model) and natural language processing system, a server-side backend, and an Android application for the frontend.
It has been awarded a selected “Super Creator” award of MITOU project 2013. (Less than 8% for all applicants, the youngest winner in time from 2000)

Supervisor: Yusuke Miyao
A technical assistant. 1) I’ve worked on making a semantic parsing system for database query generations from Japanese question sentences; 2) I’ve made a coreference resolution system as a part of a group project for a textual entailment workshop, NTCIR-10 RITE. (Our system achieves the top score on 1 out of 3 datasets among 12 teams)

Supervisor: Mamoru Komachi
A Technical Assistant. I’ve made a converter for a predicate-argument structures corpus in the Japanese language with Python.

Website: https://bitbucket.org/torotoki/corenlp-python/
A Python wrapper for Stanford CoreNLP, a set of essential libraries for natural language processing. This is a fork of Dustin Smith’s stanford-corenlp-python with new features and bugfixes.
Received new 21 forks, 29 watchers in Bitbucket; also used by natural language processing engineers/researchers, also used in a source code for an international conference paper.

Tools & dataset for predicate-argument structures and coreferences for Balanced Corpus of Contemporary Written Japanese (BCCWJ), a popular corpus for Japanese language.
This is the project when I worked at Nara Institute of Science and Technology (NAIST) with Assistant Prof. Mamoru Komachi.
Currently available as a part of BCCWJ-DepParaPAS

DG Lab Tokyo, 2016: I’ve made a contextual question answering system for experiment with PyTorch.
Gakushuin University, 2016: I’ve made a benchmark system for a classification task of legal texts with PyTorch as a freelance.
The University of Tokyo, 2023: Teaching Assistant of the lecture “Computational Complexity Theory”.

NII Today No. 60, 2013: The Younger Generation Discusses Their Hopes for the Todai Robot Project.
- Japanese: https://www.nii.ac.jp/userdata/results/pr_data/NII_Today/60/p10-11.pdf,
- English: https://www.nii.ac.jp/userdata/results/pr_data/NII_Today/60/p10-11.pdf
Languages:
- Japanese: native
- English: TOEFL iBT 86 (2020-07), TOEIC 920 (2024-10)

PLaMo-100B: A Ground-Up Language Model Designed for Japanese Proficiency, 2024. Preferred Elements et al. arXiv preprint arXiv:2410.07563. https://arxiv.org/abs/2410.07563
The Logic for a Mildly Context-Sensitive Fragment of the Lambek-Grishin Calculus, 2021. Hiroyoshi Komatsu. arXiv preprint arXiv:2101.03634. https://arxiv.org/abs/2101.03634
The Display Calculus and Proof Nets for the Lambek-Grishin Calculus, and Mildly Context-Sensitive Grammars, 2021. Hiroyoshi Komatsu. SLACS.
Latent Dirichlet Allocation for Wikipedia (Japanese), 2013. Hiroyoshi Komatsu, YANS.
BnO at NTCIR-10 RITE: A Strong Shallow Approach and an Inference-based Textual Entailment Recognition System, 2013. Ran Tian, Yusuke Miyao, Takuya Matsuzaki, Hiroyoshi Komatsu. NTCIR-10.