Yixiang Cai (蔡亦翔)

Hi! I am Yixiang Cai, an undergraduate student in Artificial Intelligence at Beijing University of Posts and Telecommunications. I will begin an M.S. program in Big Data at Peking University in 2027.

My research focuses on post-training and inference acceleration for video generation and world models. I am eager to exchange ideas and collaborate with others on research, so please feel free to contact me at caiyixiang@bupt.edu.cn.

Education

M.S. in Big Data (expected), Peking University, 2027 - 2029
B.Eng. Candidate in Artificial Intelligence, Beijing University of Posts and Telecommunications, 2023 - 2027

news

Sep 22, 2026	We released our paper, SparkDiffusion: Mitigating the High-Sparsity Trap — A Unified Framework for up to 265× Single-GPU Acceleration of Visual Generation. Explore the project page and code.
Sep 13, 2026	We released our paper, CrossDistill: Balancing Quality and Diversity via Trajectory-Level Hybrid Few-Step Distillation. Explore the code.
Sep 06, 2026	We released our paper, RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion Transformers. Explore the code.
May 20, 2026	We released our paper, RoPeSLR: 3D RoPE-driven Sparse-LowRank Attention for Efficient Diffusion Transformers.

selected publications

RoPeSLR: 3D RoPE-driven Sparse-LowRank Attention for Efficient Diffusion Transformers

Yuxi Liu^* , Zekun Zhang^* , Yixiang Cai^*, and 3 more authors

2026

Bib HTML PDF

@misc{liu2026ropeslr3dropedrivensparselowrank,
  title = {RoPeSLR: 3D RoPE-driven Sparse-LowRank Attention for Efficient Diffusion Transformers},
  author = {Liu, Yuxi and Zhang, Zekun and Cai, Yixiang and Deng, Renjia and He, Yutong and Yuan, Kun},
  year = {2026},
  eprint = {2605.20659},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2605.20659},
}

RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion Transformers

Zekun Zhang^* , Yixiang Cai^* , Yuxi Liu^*, and 9 more authors

2026

Bib HTML PDF Code

@misc{zhang2026rolarotarypositionedlowrank,
  title = {RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion Transformers},
  author = {Zhang, Zekun and Cai, Yixiang and Liu, Yuxi and Sun, Tengxu and Liu, Tianle and Wu, Zhoutong and Li, Haoyu and Ai, Baole and Wang, Ang and Wang, Jiamang and Qu, Lin and Yuan, Kun},
  year = {2026},
  eprint = {2609.06712},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2609.06712},
}

CrossDistill: Balancing Quality and Diversity via Trajectory-Level Hybrid Few-Step Distillation

Yuxi Liu^* , Haoyu Li^* , Yixiang Cai , Tengxu Sun^† , Zekun Zhang, and 6 more authors

2026

Bib HTML PDF Code

@misc{liu2026crossdistillbalancingqualitydiversity,
  title = {CrossDistill: Balancing Quality and Diversity via Trajectory-Level Hybrid Few-Step Distillation},
  author = {Liu, Yuxi and Li, Haoyu and Cai, Yixiang and Sun, Tengxu and Zhang, Zekun and Ai, Baole and Wang, Ang and Wang, Jiamang and Qu, Lin and Yuan, Kun and Zhang, Kai},
  year = {2026},
  eprint = {2609.14725},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2609.14725},
}

SparkDiffusion: Mitigating the High-Sparsity Trap — A Unified Framework for up to 265× Single-GPU Acceleration of Visual Generation

Yuxi Liu^* , Haoyu Li^* , Zekun Zhang^* , Tengxu Sun^† , Yixiang Cai, and 10 more authors

2026

Bib HTML PDF Code Website

@misc{liu2026sparkdiffusionmitigatinghighsparsitytrap,
  title = {SparkDiffusion: Mitigating the High-Sparsity Trap --- A Unified Framework for up to 265× Single-GPU Acceleration of Visual Generation},
  author = {Liu, Yuxi and Li, Haoyu and Zhang, Zekun and Sun, Tengxu and Cai, Yixiang and Li, Jiayong and Xia, Yifei and Liu, Tianle and Ai, Baole and Wang, Ang and Wang, Jiamang and Qu, Lin and Zhang, Kai and Yuan, Kun and Cui, Bin},
  year = {2026},
  eprint = {2609.23153},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2609.23153},
}