Linjia Kang
(康林佳)
ABOUT ME
Hi! I am currently a master's student at Tsinghua University, advised by
Prof. Zhi Wang. Before that, I obtained my B.Eng. degree in Data Science and Big Data Technology from Huazhong Agricultural University in 2025.
I am fortunate to be a member of the "LLM Agents & Embodied VLA" group, a small but vibrant research team made up of young, enthusiastic, and deeply committed people. Working alongside them is both a joy and an honor.
RECENT NEWS
-
[2026] New! Our work GeoMIND on spatial understanding in robotic manipulation has been accepted by NeurIPS 2026!
-
[2026/09] New! Our work Seek Before You Move on evidence seeking for vision-language navigation is available as a preprint.
-
[2026/06] New! Our work DeCoPatch on GUI grounding has been accepted to ECCV 2026! It reveals causal latent subspaces in VLMs and improves GUI grounding through decode-time causal intervention.
-
[2026/01] New! Our first work MobileGen on GUI Agent is available now! It adaptively synthesizes high-quality GUI trajectories by aligning task difficulty with the agent's capability frontier.
-
[2025/12] New! We are pleased to release our latest work Trailblazer, which positions LLMs as generalists and has been online validated through Douyin (Chinese TikTok).
-
[2025/09] New! Our first work CoBel-World on LLM multi-agent systems is available now! It introduces belief modeling for LLMs to reason like human collaborators!
RESEARCH INTERESTS
My research interests are multimodal understanding and large language model agents, particularly visual grounding, embodied navigation, and GUI interaction.
SELECTED PUBLICATION
Note: Equal contributions are marked by *.
-
DeCoPatch: Revealing Causal Latent Subspaces in Vision-Language Models for GUI Grounding
ECCV 2026 (Accepted) [THU-A]
paper
code
Yongkang Zhang*, Linjia Kang*, Zhimin Wang*, Duo Wu, Zhi Wang
-
Learning with Challenges: Adaptive Difficulty-Aware Data Generation for Mobile GUI Agent Training
Linjia Kang*, Zhimin Wang*, Yongkang Zhang*, Duo Wu, Jinghe Wang, Ming Ma, Haopeng Yan, Zhi Wang
PUBLICATIONS
-
GeoMIND: A Benchmark for Spatial Understanding in Robotic Manipulation
NeurIPS 2026 (Accepted) [CCF-A]
Jinghe Wang*, Xinrui Cao*, Duo Wu*, Chenghao Gu, Yong Zhong, Linjia Kang, Tianyi Xion, Zhi Wang.
-
DeCoPatch: Revealing Causal Latent Subspaces in Vision-Language Models for GUI Grounding
ECCV 2026 (Accepted) [THU-A]
paper
code
Yongkang Zhang*, Linjia Kang*, Zhimin Wang*, Duo Wu, Zhi Wang
-
Seek Before You Move: Evidence Seeking for Progress Grounding in Vision-Language Navigation
Zhimin Wang*, Meiyuan Zhu*, Duo Wu*, Linjia Kang, Yajun Wang, Yuan Ni, Xiaohang Wang, Tianlu Pan, Jingyan Jiang, Yaowei Wang, Zhi Wang.
-
Learning with Challenges: Adaptive Difficulty-Aware Data Generation for Mobile GUI Agent Training
Linjia Kang*, Zhimin Wang*, Yongkang Zhang*, Duo Wu, Jinghe Wang, Ming Ma, Haopeng Yan, Zhi Wang
-
Large Language Models as Generalist Policies for Network Optimization
Duo Wu, Linjia Kang, Zhimin Wang, Fangxin Wang, Wei Zhang, Xuefeng Tao, Wei Yang, Le Zhang, Peng Cui, Zhi Wang.
-
Collaborative Belief Reasoning with LLMs for Efficient Multi-Agent Collaboration
Under review [CCF-A] preprint
Zhimin Wang*, Duo Wu*, Shaokang He*, Linjia Kang, Jinhe Wang, Jing Yu, Kai Zhu, Jiawei Li, Zhi Wang.
EXPERIENCE
-
Xingchen General Artificial Intelligence Laboratory, TeleAI, Shanghai
Research Intern, Sep. 2026 – Present
-
Audio/Video Policy Model Group, Kuaishou Technology, Beijing
Research Intern, Aug. 2025 – Present
AWARDS
-
Comprehensive Excellence Scholarship (First Prize), THU SIGS, 2026
-
Outstanding Graduate, HZAU, 2025
-
National Scholarship (Top 1%), China, 2024