PeopleofSuperintelligence

Researcher

Masanori Koyama

Masanori Koyama is a machine learning researcher at the University of Tokyo and a coauthor of spectral normalization for GANs and of the Optuna hyperparameter optimization framework. He was in the Department of Mathematics at the University of Wisconsin–Madison in 2013 and in the Department of Mathematical Science at Ritsumeikan University in 2017. In 2018, he coauthored Spectral Normalization for Generative Adversarial Networks and cGANs with Projection Discriminator, both published at ICLR 2018, and in 2019 he coauthored Optuna at Preferred Networks. His later papers cover equivariant representation learning (Neural Fourier Transform, 2024), flow matching theory (2025) and block-wise training of transformers (DiffusionBlocks, 2026). He is a project lecturer in the Graduate School of Engineering at the University of Tokyo.

Badges

Timeline

  1. 2013

    Mathematics at UW–Madison

    He is in the Department of Mathematics at the University of Wisconsin–Madison in 2013, when he coauthors a paper in Genetics, Selection, Evolution with Gota Morota, Guilherme J. M. Rosa, Kent A. Weigel and Daniel Gianola. The paper applies a diffusion kernel on genetic markers to predict complex traits in dairy cattle and wheat data.

  2. 2017

    Ritsumeikan and reaction networks

    In August 2017, he posts a paper with David F. Anderson, Daniele Cappelletti and Thomas G. Kurtz that proves complex-balanced stochastic reaction networks are non-explosive. His affiliation on the paper is the Department of Mathematical Science at Ritsumeikan University.

  3. 2018

    Spectral normalization for GANs

    In 2018, he coauthors Spectral Normalization for Generative Adversarial Networks with Takeru Miyato, Toshiki Kataoka and Yuichi Yoshida; the weight normalization technique stabilizes the training of the discriminator. With Miyato he also writes cGANs with Projection Discriminator, which feeds conditional information into the discriminator by projection. Both papers are published at ICLR 2018.

  4. 2019

    Optuna at Preferred Networks

    Listed at Preferred Networks, Inc., he coauthors Optuna with Takuya Akiba, Shotaro Sano, Toshihiko Yanase and Takeru Ohta. The hyperparameter optimization framework is built around a define-by-run API that builds the search space dynamically; the paper is accepted at KDD 2019 and the software is released under the MIT license.

  5. 2024

    Neural Fourier Transform

    In 2024, he is first author of Neural Fourier Transform, written with Kenji Fukumizu, Kohei Hayashi and Takeru Miyato and published at ICLR 2024. The framework learns the latent linear action of a group without assuming explicit knowledge of how the group acts on data.

  6. 2024–2025

    Flow matching convergence theory

    In October 2024, he is at Preferred Networks and revises, with Kenji Fukumizu, Taiji Suzuki, Noboru Isobe and Kazusato Oko, a paper on flow matching that is published at ICLR 2025. The paper shows that flow matching, which generates samples by solving an ordinary differential equation from a normal distribution, achieves an almost minimax optimal convergence rate.

  7. 2026

    DiffusionBlocks at ICLR

    In June 2026, he is at the University of Tokyo, coauthoring DiffusionBlocks with Makoto Shing and Takuya Akiba of Sakana AI. The ICLR 2026 paper turns transformer networks into independently trainable blocks through a denoising process, which reduces memory requirements in proportion to the number of blocks.

  8. As of October 2026

    Project lecturer at Tokyo

    He is a project lecturer in the University of Tokyo's Graduate School of Engineering. In 2026, he coauthors C-Voting with Kenji Kubo, Shunsuke Kamiya, Kohei Hayashi, Yusuke Iwasawa and Yutaka Matsuo, a test-time scaling strategy for recurrent reasoning models published at ICLR 2026.

Career

  1. NowUniversity of TokyoProject Lecturer
  2. From 2019Preferred Networks
  3. 2017Ritsumeikan University
  4. 2013University of Wisconsin–Madison

From the biography.

  1. People shaping AI work at a median of 3 organizations over a career: 3 for university starters and institute and lab starters, versus 2 for company startersIncludes University of Tokyo. Across 6,853 people shaping AI with a dated work role, we estimate a median of 3 parent organizations over a career (mean 3.6), comparing role groups, starting sectors and starting decades from the 1950s to 2020–2026.
  2. Universities keep AI researchers for a median of 6 years, compared with 2 years at frontier labsIncludes University of Tokyo. Across 15,811 dated work stays (6,529 people), median role tenure runs from 6 years at universities to 2 years at frontier labs. We compare all five organization types, starting decades from the 1970s to the 2020s, and the organizations that keep people longest.
  3. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes University of Tokyo. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  4. The United States employs an estimated 48% of people shaping AIIncludes University of Tokyo. We estimate where the people shaping AI work today, how universities, companies and research institutes split across countries, and how country shares have changed since the 1960s.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Masanori Koyama yet. Send us a free-licensed photo.

  1. Predicting complex traits using a diffusion kernel on genetic markers with an application to dairy cattle and wheat data - PMC pmc.ncbi.nlm.nih.gov. Checked 21 October 2026.
  2. [1708.09356] Non-explosivity of stochastically modeled reaction networks that are complex balanced arxiv.org. Checked 21 October 2026.
  3. [1708.09356] Non-explosivity of stochastically modeled reaction networks that are complex balanced (full text) ar5iv.labs.arxiv.org. Checked 21 October 2026.
  4. [1802.05957] Spectral Normalization for Generative Adversarial Networks arxiv.org. Checked 21 October 2026.
  5. [1802.05637] cGANs with Projection Discriminator arxiv.org. Checked 21 October 2026.
  6. [1907.10902] Optuna: A Next-generation Hyperparameter Optimization Framework arxiv.org. Checked 21 October 2026.
  7. [1907.10902] Optuna: A Next-generation Hyperparameter Optimization Framework (full text) ar5iv.labs.arxiv.org. Checked 21 October 2026.
  8. Neural Fourier Transform: A General Approach to Equivariant Representation Learning arxiv.org. Checked 21 October 2026.
  9. Koyama, Masanori | ML Anthology mlanthology.org. Checked 21 October 2026.
  10. Flow matching achieves almost minimax optimal convergence arxiv.org. Checked 21 October 2026.
  11. DiffusionBlocks: Block-wise Neural Network Training via Diffusion Interpretation arxiv.org. Checked 21 October 2026.
  12. C-Voting: Confidence-Based Test-Time Voting without Explicit Energy Functions arxiv.org. Checked 21 October 2026.
  13. KOYAMA Masanori | The University of Tokyo u-tokyo.ac.jp. Checked 21 October 2026.