Linhan Wang

I'm a Ph.D. student in Computer Science at Virginia Tech, advised by Prof. Chang-Tien Lu. I most recently spent time at XPENG as a research intern working on end-to-end autonomous driving.

Before starting my Ph.D. I worked for about five years as a software and machine learning engineer — building ML platforms at Didi and Weibo, and low-latency trading systems at a quantitative fund in Beijing. I received my B.S. in Atmospheric and Oceanic Science from the School of Physics at Peking University.

My research is on embodied AI and world models for self-driving and robotics. Along the way I've also worked on neural rendering and 3D Gaussian splatting, and on label-efficient learning for scientific and archival imagery.

Linhan Wang

News

Selected Publications

* denotes work under review. For the full list, see my Google Scholar or CV.

GlanceWAM sparse imagination overview

GlanceWAM: Sparse Test-Time Imagination for World-Action Models

Linhan Wang, Zijian An, Mingyuan Zhang, Chen Dai, Yi Xu, Can Cui, Zichong Yang, Yinlin Chen, Lifeng Zhou, Chang-Tien Lu

Under review, 2026 *

arXiv / Code

Decouples asynchronous background video imagination from a latency-free action decoder inside one video diffusion transformer — 72.2% success on RoboCasa and 99.0% on LIBERO, 24x faster than synchronous baselines at 48ms per action chunk.

Drive-JEPA architecture overview

Drive-JEPA: Video JEPA Meets Multimodal Trajectory Distillation for End-to-End Driving

Linhan Wang, Zichong Yang, Guoxiang Zhang, Xiaotong Liu, Xiaoyin Zheng, Xiao-Xiao Long, Chang-Tien Lu, Chen Bai

Under review, 2026 *

arXiv / Code / Dataset

Self-supervised V-JEPA pretraining on curated driving video, plus simulator-scored pseudo-teacher trajectories: 93.3 PDMS on NAVSIM v1 and 87.8 EPDMS on v2.

Recall-oriented VLM tagging pipeline

Learning Subject Tags from Incomplete Labels: A Recall-Oriented VLM Pipeline for Historical Photographs

Linhan Wang, Meizi Song, Shengkun Wang, Chang-Tien Lu, Yinlin Chen

ACM/IEEE Joint Conference on Digital Libraries (JCDL), 2026, Short Paper

Order-invariant SFT, recall-oriented GRPO, and test-time self-consistency turn missing-not-at-random curator tags into a usable supervision signal for archival photo cataloging.

SemiETPicker teacher-student framework

SemiETPicker: Fast and Label-Efficient Particle Picking for CryoET Tomography Using Semi-Supervised Learning

Linhan Wang, Jianwen Dou, Wang Li, Shengkun Wang, Zhiwu Xie, Chang-Tien Lu, Yinlin Chen

IEEE International Symposium on Biomedical Imaging (ISBI), 2026, Oral

arXiv

Timber origin prediction pipeline

Chasing the Timber Trail: Machine Learning to Reveal Harvest Location Misrepresentation

Shailik Sarkar, Raquib Bin Yousuf, Linhan Wang, Brian Mayer, Thomas Mortier, Victor Deklerck, Jakub Truszkowski, John C. Simeone, Marigold Norman, Jade Saunders, Chang-Tien Lu, Naren Ramakrishnan

ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2025

arXiv / ACM DL

KHAIT urban search and rescue site

KHAIT: K-9 Handler Artificial Intelligence Teaming for Collaborative Sensemaking

Matthew Wilchek, Linhan Wang, Sally Dickinson, Erica Feuerbacher, Kurt Luther, Feras A. Batarseh

ACM Conference on Intelligent User Interfaces (IUI), 2025

arXiv / ACM DL

DC-Gaussian dash cam reflection removal

DC-Gaussian: Improving 3D Gaussian Splatting for Reflective Dash Cam Videos

Linhan Wang, Kai Cheng, Shuo Lei, Shengkun Wang, Wei Yin, Chenyang Lei, Xiaoxiao Long, Chang-Tien Lu

Conference on Neural Information Processing Systems (NeurIPS), 2024

Project page / arXiv / Code

SCCNet self- and cross-correlation pipeline

Self-Correlation and Cross-Correlation Learning for Few-Shot Remote Sensing Image Semantic Segmentation

Linhan Wang, Shuo Lei, Jianfeng He, Shengkun Wang, Min Zhang, Chang-Tien Lu

ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SIGSPATIAL), 2023

arXiv / ACM DL / Code

Experience

Education

Awards