Guang Li

Assistant Professor at Hokkaido University.
Working on dataset distillation, data-efficient learning,
and generative modeling.

guang (at) lmd.ist.hokudai.ac.jp

About Me

Hi, my name is Guang Li. I am an Assistant Professor at the Education and Research Center for Mathematical and Data Science, Hokkaido University, and a member of the Laboratory of Media Dynamics.

I received dual bachelor's degrees in Software Engineering and Japanese from Dalian University of Technology in 2019, and my M.S. and Ph.D. degrees in Information Science from Hokkaido University in 2022 and 2023, under the supervision of Prof. Miki Haseyama and Prof. Takahiro Ogawa. I collaborate closely with Prof. Konstantinos N. Plataniotis at the University of Toronto.

I have been contributing to the field of dataset distillation since its early days, maintain Awesome-Dataset-Distillation, and co-organized the first CVPR Workshop on Dataset Distillation and the first Dataset Distillation Challenge at ECCV. I received the 2024 IEEE Sapporo Young Professionals Best Researcher Award and the ICML 2026 Gold Reviewer Award, and serve as an Area Chair for ACM MM, MICCAI, and ICASSP. My research has been covered by Scientific American, Nikkei, and NHK.

Experience

Hokkaido University Assistant Professor, Education and Research Center for Mathematical and Data Science Oct. 2023 – present

Education

Hokkaido University Ph.D., Information Science Apr. 2022 – Sep. 2023
Hokkaido University M.S., Information Science Apr. 2020 – Mar. 2022
Dalian University of Technology B.S., Software Engineering & B.A., Japanese Sep. 2015 – Jun. 2019

News

Publications

Projects

GITHUB · CURATED LIST Awesome Dataset Distillation

The community-standard curated list of dataset distillation papers, code, and resources — maintained since 2022 with Bo Zhao and Tongzhou Wang.

GITHUB · COMMUNITY BENCHMARK DD-Ranking

A fair, unified benchmark for evaluating dataset distillation methods, built with the NUS HPC-AI Lab and the dataset distillation community.

GITHUB · RESEARCH CODE RC-Aux for Latent World Models

Predictive but not plannable: reward-consistency auxiliary objectives for latent world models. Code, paper, and project page.

GITHUB · RESEARCH CODE FD²: Fine-Grained Dataset Distillation

Official implementation of FD² (ECCV 2026) — counterfactual attention learning and prototype constraints for fine-grained dataset distillation.

GITHUB · RESEARCH CODE EVLF: Early Vision-Language Fusion

Official implementation of EVLF (CVPR 2026) — fusing image and text embeddings early via cross-attention for generative dataset distillation.

GITHUB · RESEARCH CODE ASMIL: Attention-Stabilized MIL

Official implementation of ASMIL (ICLR 2026) — stabilizing attention in multiple instance learning for whole-slide pathology imaging.

GITHUB · RESEARCH CODE Hyperbolic Dataset Distillation

Official implementation of HDD (NeurIPS 2025) — distribution matching in Lorentz hyperbolic space to preserve hierarchical structure in distilled data.