About me
I am a PhD student at College of Computing and Data Science, Nanyang Technological University under the supervision of Prof. Jaehong Yoon. I graduated from The University of Tokyo in March 2026. I am currently undertaking a part-time research attachment at the Institute for Infocomm Research, A*STAR I2R, under the supervision of Dr. Xun Xu.
My research interests center on discrete and continuous diffusion models and their applications, including image generation, language modeling, 3D generation, and text-to-speech systems. More broadly, I am interested in generative modeling, computer vision, 3D world understanding, and statistical machine learning.
Outside research, I enjoy playing badminton. I have brought my badminton racket to every country I have visited, including Sweden, Saudi Arabia, Switzerland, and Singapore.
News
OCT2026
Our paper What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models appeared on arXiv. GitHub · Hugging Face
SEP2026
AUG2026
Our paper Simplex Relaxation for Discrete Diffusion appeared on arXiv.
AUG2026
Our paper Test-Time Scaling for Safe Text-Guided Image Generation via Intermediate Clean Estimates appeared on arXiv.
AUG2026
I started my PhD at Nanyang Technological University under the supervision of Prof. Jaehong Yoon!
JUL2026
Our paper FairT2I was accepted to Transactions on Machine Learning Research (TMLR)! Proceedings
MAY2026
MAR2026
I graduated from The University of Tokyo.
NOV2025
I started visiting the Institute for Infocomm Research, A*STAR I2R, under the supervision of Dr. Xun Xu.
FEB2025
Our paper FairT2I: Mitigating Social Bias in Text-to-Image Generation via Large Language Model-Assisted Detection and Attribute Rebalancing appeared on arXiv.
Research Experience
Institute for Infocomm Research, A*STAR I2R
Visiting Student with Dr. Xun Xu
Nov. 2025 – May 2026
EPFL
Summer@EPFL with Prof. Amir Zamir
Jul. 2024 – Nov. 2025
- Developing a transformer pretrained model for 3D modalities using masked token prediction.
- Training VQ-VAE-style tokenizers for LRM-triplane and Gaussian splatter images.
- This work led to OmniDiMM, accepted to NeurIPS 2026.
KAUST
Visiting Student Research Program with Prof. Bernard Ghanem
Sep. 2023 – Mar. 2024
- Developed an efficient token-dropping method for VideoMAE using motion vectors from H.264 compression to reduce computation and GPU memory usage.
Albert Inc.
Research Intern
Aug. 2021 – Jul. 2022
- Developed Surface Reconstruction from Raw Point Cloud via Energy-Based Models, improving robustness to noisy point clouds.
Publications
What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models
Jinya Sakurai, Xun Xu, Jaehong Yoon
Simplex Relaxation for Discrete Diffusion
Jinya Sakurai, Patrick Pynadath, Satoshi Hayakawa, Jaehong Yoon, Xulei Yang, Nancy F. Chen, Xun Xu
OmniDiMM: Bridging 1D, 2D, and 3D with Any-to-Any Multimodal Modeling
Jason Toskov, Oriol Barbany, Rishubh Singh*, Jinya Sakurai, et al.
Test-Time Scaling for Safe Text-Guided Image Generation via Intermediate Clean Estimates
Jinya Sakurai, Shueicheng Yan, Xun Xu
FairT2I: Latent Variable Guidance for Training-Free Bias Mitigation with LLM-Assisted Bias Detection
Jinya Sakurai, Yuki Koyama, Issei Sato
Education
Nanyang Technological University
PhD in Computer Science
Aug. 2026 – Present
The University of Tokyo
Master of Computer Science
Apr. 2024 – Mar. 2026
The University of Tokyo
Bachelor of Information Science
Apr. 2019 – Sep. 2023
