Kaiwen Xue

Hello, I am

Kaiwen Xue 薛凯文

Ph.D. Candidate · Multimodal & Embodied Intelligence

Beijing University of Posts and Telecommunications
Zhongguancun Laboratory

I am a jointly trained Ph.D. candidate working on multimodal large language models, embodied geo-localization, and world models. I am interested in how intelligent agents can actively seek visual evidence, reason about space, and make reliable decisions in open environments.

我是北京邮电大学与中关村国家实验室联合培养博士生，研究方向包括多模态大模型、具身地理定位与世界模型。

Research interests: Multimodal LLMs · Embodied Geo-localization · World Models · Creativity Evaluation
EmailGoogle ScholarGitHub

Background

Experiences

  1. Beijing, China

    Beijing University of Posts and Telecommunications

    Ph.D. Candidate in Cyberspace Security

    Jointly trained with Zhongguancun Laboratory. My research focuses on multimodal large language models, embodied geo-localization, and world models.

  2. Guangzhou, China

    South China Normal University

    M.Eng. in Electronic Information

    Worked on cell-nucleus segmentation and intelligent analysis for digital pathology images.

  3. Taiyuan, China

    North University of China

    B.Eng. in Optoelectronic Information Science and Engineering

    Conducted undergraduate research in digital holography and acoustic levitation.

Research in practice

Selected Projects

National Key R&D Program of China

Creativity evaluation for key technologies in intelligent interactive experimental teaching.

Beijing Institute of Control Engineering

Multimodal perception and autonomous agent workflows for space-target understanding.

China Mobile Research Project

Led research and delivery of a meta-learning system for network-state diagnosis.

AI-assisted Pathological Grading

Led a student research project on deep-learning-based grading of breast-cancer biopsy slides.

Updates

News

  1. ERGeoBench was accepted by ICML 2026.

  2. CreBench was published at AAAI 2026.

  3. Selected as an Outstanding Graduate Student, top 5% of the program.

  4. Received the First Prize of the Science and Technology Progress Award from the China Association of Work Safety.

Research outputs

Selected Publications

Full list on Google Scholar
ICML2026First author · CCF-A

ERGeoBench: A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models

Kaiwen Xue, Tao Wei, Guoxin Zhang, Zhonghong Ou, Kaoyan Lu, Yu Feng, Yifan Zhu, Haoran Luo

A benchmark spanning single-view, panorama-view, and embodied exploration over 2,207 globally distributed street panoramas.

Project Paper Code
AAAI2026Co-first author · CCF-A

CreBench: Human-Aligned Creativity Evaluation from Idea to Process to Product

Kaiwen Xue*, Chenglong Li*, Zhonghong Ou, Guoxin Zhang, Kaoyan Lu, Shuai Lyu, et al.

A human-aligned benchmark and 4.7M-instruction dataset for evaluating multimodal creativity across ideas, processes, and products.

Project Paper Code
Applied Sciences2024Co-first author · SCI Q1

LogEDL: Log Anomaly Detection via Evidential Deep Learning

Yunfeng Duan*, Kaiwen Xue*, Hao Sun, Haotong Bao, Yadong Wei, et al.

An evidential learning approach that quantifies predictive uncertainty to detect unseen log anomalies in open settings.

Paper
Under review2027First author

LaGeo: Latent Evidence Search with Spatial Belief for Embodied Geolocation

Kaiwen Xue et al.

A latent evidence-search framework for informative viewpoint selection and pose-aware spatial belief.

Recognition

Selected Honors