Taiji Suzuki
Taiji Suzuki develops mathematical theory for machine learning. His RIKEN team studies how complex models learn and how to optimize them efficiently. Suzuki and Atsushi Nitanda coauthored a study of averaged stochastic gradient descent in the neural tangent kernel regime. The work received an ICLR 2021 outstanding paper award. With Kazusato Oko and Shunta Akiyama, Suzuki coauthored an ICML paper on diffusion models as minimax-optimal distribution estimators. This connects generative modeling to statistical estimation theory.
- Now at
- RIKEN and University of Tokyo
- Works
- 278
- Citations
- 4,056
- h-index
- 32
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
2021
Neural-network optimization
Suzuki and Atsushi Nitanda coauthored a study of averaged stochastic gradient descent in the neural tangent kernel regime. The work received an ICLR 2021 outstanding paper award.
-
2023
Theory of diffusion models
With Kazusato Oko and Shunta Akiyama, Suzuki coauthored an ICML paper on diffusion models as minimax-optimal distribution estimators. This connects generative modeling to statistical estimation theory.
-
2024
Features learned by transformers
Suzuki and Juno Kim coauthored an ICML paper on how transformers learn nonlinear features in context. The research examines learning dynamics in the attention mechanism.
-
As of October 2026
Deep Learning Theory Team
Suzuki directs RIKEN’s Deep Learning Theory Team. The team combines statistical learning theory, high-dimensional statistics and stochastic optimization.
Sources and credits
Photo: www.riken.jp; creator not stated. License: Not stated by source. Source: riken.jp. Cropped to a square and resized. Ask us to remove or replace this photo.