Jingyi Liu (刘璟奕)

I am a year-0 Ph.D. student at the Camera Intelligence Lab, Peking University, advised by Prof. Boxin Shi. My research focuses on video generation and world models.

I am currently working on autoregressive video generation.

I welcome research discussions and collaborations—please feel free to reach out via email. Outside research, I enjoy swimming and photography.

Email GitHub Scholar 知乎

Research

My research interests include world models, video generation, multimodal learning, and representation learning.

News

Sep 2026I joined BAAI as a World Model Research Intern.
Jun 2026One paper was accepted to ECCV 2026.
Jun 2026I will serve as a reviewer for AAAI 2027.
May 2026One paper was accepted to ICML 2026.

More news

Feb 2026One paper was accepted to CVPR 2026.
Jan 2026One paper was accepted to ICASSP 2026 (Oral).
Nov 2025I was awarded the Fan Linhu Petroleum Scholarship.
Oct 2025I was awarded the National Scholarship.
Oct 2025I served as a reviewer for CVPR 2026.
Sep 2025I served as a reviewer for ICASSP 2026.
Jul 2025I served as a reviewer for AAAI 2026.

Publications

Mixture of Specialized Vision Experts: Unlocking Complementary Visual Insights for Faithful MLLM Reasoning

Yifei Gao, Liangliang You, Jiye Xie, Changwei Wang, Kexue Fu, Jingyi Liu, Rongtao Xu, Zhiqiang Kou, Haoran Xu, Longxiang Gao, Yu Zhang

ECCV 2026

MoSVE combines complementary visual experts to reduce hallucinations and improve visually grounded reasoning in multimodal large language models.

Adaptive Visual Autoregressive Acceleration via Dual-Linkage Entropy Analysis

Yu Zhang, Jingyi Liu^†, Feng Liu, Duoqian Miao^*, Qi Zhang, Kexue Fu, Changwei Wang, Longbing Cao

ICML 2026

† Project Lead · * Corresponding Author

Paper

NOVA accelerates visual autoregressive generation without additional training, using entropy-guided token reduction across scales and layers together with residual caching.

Markovian Scale Prediction: A New Era of Visual Autoregressive Generation

Yu Zhang, Jingyi Liu^†, Yiwei Shi, Qi Zhang, Duoqian Miao^*, Changwei Wang, Longbing Cao

CVPR 2026

† Project Lead · * Corresponding Author

Paper Project

Markov-VAR reformulates scale prediction as a Markov process, using a compact history vector to reduce reliance on full context, lowering memory usage while further improving generation quality.

Enhancing Quantization for Visual AutoRegressive Generation via Uncertainty Identification

Jingyi Liu, Guangyao Xiong, Shuxia Ren^†, Ruoyang Gao, Linkun Yu, Kexue Fu

ICASSP 2026 · Oral

† Corresponding Author

This work explores uncertainty-aware quantization for visual autoregressive generation, using entropy to identify uncertain prediction scales.

Education

Peking University · Camera Intelligence Lab

Year-0 Ph.D. student · advised by Prof. Boxin Shi

Program start: September 2027

Tongji University · School of Ocean and Earth Science

Undergraduate · Marine Technology major, Computer Science and Technology minor

September 2023 – Present

Internships

Beijing Academy of Artificial Intelligence (BAAI)

World Model Research Intern

September 2026 – Present

YITU, Shanghai

Algorithm Intern

July 2025 – October 2025

Honors and Awards

National Scholarship (2025)
Fan Linhu Petroleum Scholarship (2025, Top 2%)
Tongji University Geophysics Scholarship (2024, Top 5%)
First Prize, National Finals of Global Campus AI Algorithm Competition (2024)
Second Prize, National AI Application Innovation Challenge (2024)
Second Prize, 15th National College Mathematics Competition (2023)