Siyong Jian

Hi! I am currently working with Prof. Xin Jin and Prof. Wenjun Zeng at CVLab, EIT. Previously, I was fortunate to work with with Prof. Huan Wang and Siyuan Li at ENCODE-Lab, Westlake University. I received my M.S. from Nanjing University and my B.S. from Zhengzhou University. My research focuses on efficient and embodied AI.

Research Interests: Efficient Generative Models, Embodied AI.

Email | WeChat | LinkedIn | GitHub | Scholar
Portrait of Siyong Jian

Publications & Preprints

* equal contribution    † corresponding author

Teaser for RankE

RankE: End-to-End Post-Training for Discrete Text-to-Image Generation with Decoder Co-Evolution

Siyong Jian*, Siyuan Li*, Luyuan Zhang*, Zedong Wang, Xin Jin, Ying Li, Cheng Tan†, Huan Wang†

arXiv'26 | arXiv | Code

RankE presents an end-to-end post-training framework for discrete text-to-image generation, coupling generator and decoder optimization through decoder co-evolution and reward-driven training.

Teaser for Parallel Jacobi Decoding

Parallel Jacobi Decoding for Fast Autoregressive Image Generation

Boya Liao, Ying Li, Siyong Jian, Huan Wang

CVPR'26 | Paper coming soon

A training-free decoding method that refines draft tokens in the 2D spatial domain to accelerate autoregressive image generation while preserving visual fidelity.

Teaser for MergeMix

MergeMix: A Unified Augmentation Paradigm for Visual and Multi-Modal Understanding

Xin Jin*, Siyuan Li*, Siyong Jian, Kai Yu, Huan Wang

ICLR'26 | arXiv | Code

MergeMix unifies visual augmentation and preference alignment by constructing context-aware mixed samples for both visual and multi-modal understanding.

Teaser for SSD project

SSD: Spatial-Semantic Head Decoupling for Efficient Autoregressive Image Generation

Siyong Jian, Huan Wang

arXiv'25 | arXiv | Code

A lightweight autoregressive image generation project that decouples spatial and semantic prediction heads to improve efficiency while preserving generation quality.

Teaser for ERC-SVD

ERC-SVD: Error-Controlled SVD for Large Language Model Compression

Haolei Bai, Siyong Jian, Tuo Liang, Yu Yin, Huan Wang

CPAL'26 | arXiv

ERC-SVD explores efficient representation compression and acceleration with the updated teaser and conference information you provided.