Seiya Tokui is a researcher at Preferred Networks. He received the doctor’s degree in information science and technology at the University of Tokyo in 2022. He was the lead developer of a deep learning framework, Chainer. His current interests in research and development include deep learning and its software stack.

Education

Doctor: Computer Science, U. Tokyo (Apr. 2016 – Sept. 2022)

  • Research theme: deep variational models, disentangled representation learning, gradient estimation for stochastic computation
  • Thesis: Learning and Evaluating Deep Generative Models for Disentanglement
  • Supervisor: Issei Sato

Master: Mathematical Informatics, U. Tokyo (Apr. 2010 – Mar. 2012)

  • Research theme: machine learning, nearest neighbor search, natural language processing
  • Master’s thesis: Learning Hash with Sequential Buckets Partitioning
  • Supervisor: Hiroshi Nakagawa

Bachelor: Mathematics, U. Tokyo (Apr. 2006 – Mar. 2010)

  • Studied topology and combinatorial optimization

Job Experiences

Researcher at Preferred Networks (Oct. 2014 – current)

  • Worked on machine learning and computer vision for IoT
  • Led the development of Chainer, a framework for deep learning
  • Working on a part of the deep learning software stack

Researcher/Engineer at Preferred Infrastructure (Apr. 2012 – Oct. 2014)

  • Worked on machine learning, natural language processing, and computer vision
  • Was mainly engaged in the Jubatus project

Internship

  • pixiv (Mar. 2011), worked on improvements of recommender systems
  • Google Japan (Aug. 2010 – Sep. 2010), worked at Google Japanese Input team (a.k.a. mozc) and created the calculator feature

Publications and Presentations

Conference Papers (refereed)

Seiya Tokui, Issei Sato. Disentanglement Analysis with Partial Information Decomposition. 10th International Conference on Learning Representations (ICLR), 2022. [OpenReview, arXiv]

Seiya Tokui, Ryosuke Okuta, Takuya Akiba, Yusuke Niitani, Toru Ogawa, Shunta Saito, Shuji Suzuki, Kota Uenishi, Brian Vogel and Hiroyuki Yamazaki Vincent. Chainer: a Deep Learning Framework for Accelerating the Research Cycle. 25th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2019.

Seiya Tokui, Issei Sato. Evaluating the Variance of Likelihood-Ratio Gradient Estimators. International Conference on Machine Learning (ICML), 2017. [PMLR]

Weihua Hu, Takeru Miyato, Seiya Tokui, Eiichi Matsumoto, Masashi Sugiyama. Learning Discrete Representations via Information Maximizing Self-Augmented Training. International Conference on Machine Learning (ICML), 2017. [arXiv]

Seiya Tokui, Issei Sato, Hiroshi Nakagawa. Locally Optimized Hashing for Nearest Neighbor Search. In Advances in Knowledge Discovery and Data Mining, 19th Paci c-Asia Conference, PAKDD 2015, Ho Chi Minh City, Vietnam, May 19-22, 2015, Proceedings, Part II, 498–509.

Workshop Papers (refereed)

Seiya Tokui, Kenta Oono, Shohei Hido, Justin Clayton. Chainer: a Next-Generation Open Source Framework for Deep Learning. In Workshop on Machine Learning Systems at Neural Information Processing Systems (NIPS), 2015.

Shohei Hido, Satoshi Oda and Seiya Tokui. Jubatus: An Open Source Platform for Distributed Online Machine Learning. In Big Learning Workshop at Neural Information Processing Systems (NIPS), 2013.

Tutorials

Seiya Tokui, Kenta Oono, Atsunori Kanemura. Deep Learning Implementations and Frameworks, at the Thirty-first Conference on Artificial Intelligence (AAAI), 2017.

Seiya Tokui, Kenta Oono, Atsunori Kanemura, and Toshihiro Kamishima. Deep Learning Implementations and Frameworks, at the 20th Pacific Asia Conference on Knowledge Discovery and Data Mining (PAKDD), 2016.

Others

  • I have written a Japanese book on online machine learning with three of my colleagues.
  • See Japanese version for the domestic conferences and workshops.

Awards

Seiya Tokui. Software Japan Award 2017 (for the development of a deep learning framework). Information Processing Society of Japan, 2017.

Software Projects

Chainer: a Framework for Deep Learning (Apr. 2015 – current)

maf: a Build Tool for Parameterized Experiments (Dec. 2012 – Mar. 2015)

Jubatus: a Distributed Machine Learning Framework (Apr. 2012 – Apr. 2014)

Skill

  • Programming: C++/Python (advanced), CUDA, Rust
  • Communication: Japanese (native), English