I am a Research Scientist at ByteDance in San Jose, working on world modeling and generative AI.
I received my Ph.D. in Electrical and Computer Engineering from Duke University in 2026, advised by Prof. Guillermo Sapiro, and spent the last two years of my Ph.D. as a visiting student at Princeton University. Before that, I received my M.S. from Peking University and my B.E. from Huazhong University of Science and Technology.
My work spans video generation and editing, visual autoregressive modeling, and representation learning, with papers at ICCV (Oral), NeurIPS, ICLR, and CVPR workshops. Along the way I have worked at ByteDance, Microsoft Research Asia, and Tencent AI Lab.
A multimodal Semantic Planner reasons over text and visual context and writes spatio-temporal semantic tokens that guide a video diffusion model, which cuts visual hallucinations and makes multi-step instructions land.
Animates a reference image from any driving video's motion, camera, effects, and shot transitions, using self-supervision on unpaired in-the-wild videos.
Makes orthogonal low-rank embedding work for SSL by decoupling low- and high-rank objectives and constraining deviations from the mean. It needs no large batches, memory banks, or dual encoders.
A survey of pixel-, token-, and scale-based visual autoregressive models across image, video, 3D, and multimodal generation, with about 250 references.
Progressive residual quantization learns a coarse-to-fine sequence of discrete codes that carries as much information as possible within a fixed description length.
An early open-domain text-to-video model. It generates VQ-VAE tokens autoregressively with 3D sparse attention, pretrained on HowTo100M.
An information vector and attention gate filter what attention returns. AoANet set a new state of the art of 129.8 CIDEr-D on MS COCO.
Lets the decoder learn how many attention steps to take for each word, so regions and words align many-to-many. The method is deterministic and differentiable.
* Equal contribution. The full list is on Google Scholar.