Earth observation · Multimodal AI · Scientific discovery

Learning to understand
our Earth.

I develop foundation models that connect Earth observation, multimodal intelligence, and scientific discovery.

I am a Postdoctoral Research Scientist at Singapore Management University, advised by Prof. Qianru Sun. I received my Ph.D. in Computer Science and Technology from Nankai University in 2026, advised by Professors Jian Yang, Xiang Li, and Ming-Ming Cheng. Before that, I completed a four-year integrated M.Eng. in Computer Science at University College London with First-Class Honours.

My work spans multi-modal remote sensing perception, efficient specialization and continual learning of multimodal large language models (MLLMs), and self-evolving research agents for Earth and environmental science. I welcome research collaborations; please get in touch by email.

Earth observation

Foundation models and unified perception across SAR, optical, and infrared imagery.

Multimodal learning

Fast domain specialization, post-training, and continual adaptation of open MLLMs.

AI for science

Self-evolving research agents for Earth and environmental discovery.

2,712Google Scholar citations
13h-index
2,200+GitHub stars across projects

Statistics as of November 2026. View Google Scholar for current citation counts.

Selected publications

Full list on Google Scholar

Selected first- and second-author works.

Under review

Figure 1 from HOPD: Historical On-Policy Distillation for SAR MLLM Continual Learning
Under reviewNov 2026 · First author

HOPD: Historical On-Policy Distillation for SAR MLLM Continual Learning

Yuxuan Li, Derrick Goh Xin Deik, Zheng-Peng Duan, Qianru Sun, and Weijie Li

Historical on-policy distillation for continual SAR visual question answering, using source-acquisition LoRA states as persistent teachers to adapt MLLMs across evolving sensors and regions while mitigating catastrophic forgetting.

GitHub stars Google Scholar citations

Conference & journal papers

Paper illustration from LSKNet: A Foundation Lightweight Backbone for Remote Sensing
IJCV 2025 ESI Highly Cited PaperFirst author

LSKNet: A Foundation Lightweight Backbone for Remote Sensing

Yuxuan Li, Xiang Li, Yimian Dai, Qibin Hou, Li Liu, Yongxiang Liu, Ming-Ming Cheng, and Jian Yang

A lightweight remote sensing backbone with dynamically adjustable large spatial receptive fields to capture contextual dependencies across geographic scenes.

GitHub stars Google Scholar citations
Additional publications & collaborations

Appointments & education

Sep 2026 – Jun 2027

Singapore Management University

Postdoctoral Research Scientist · Singapore

Remote sensing foundation models, multimodal LLMs, and AI agents.
Advisor: Qianru Sun

Sep 2023 – Jun 2026

Nankai University

Ph.D. in Computer Science and Technology · Tianjin, China

Computer vision. Advisors: Jian Yang, Xiang Li, and Ming-Ming Cheng.

2021 – 2023

Nankai University

Research Assistant · Tianjin, China

Sep 2017 – Jun 2021

University College London

Integrated Bachelor & Master (M.Eng.) in Computer Science · London, UK

First-Class Honours · Four-year programme

Selected honors & awards

  • 2026
    World’s Top 2% Scientists

    Stanford University & Elsevier · Single-Year Impact

  • 2025
    National Scholarship for Ph.D. Students

    China · ¥30,000

  • 2024, 2025
    GongNeng Scholarship, First Class

    Nankai University

  • 2023
    Champion, 2nd Jittor AI Global Challenge

    Team leader · 1st of 154 teams · ¥50,000 · Code

  • 2022
    Runner-up, Greater Bay Area International Algorithm Competition

    Team leader · ¥100,000 · Code

  • 2019
    Runner-up, Facebook Hack-a-Project Spring

    London

Academic service

Conference reviewer
NeurIPS, ICML, ICCV, CVPR, ECCV, AAAI
Journal reviewer
IEEE TPAMI, IJCV, IEEE TGRS, IEEE TCSVT

Technical expertise

Foundation models & agents. MLLM post-training, LoRA, on-policy distillation, continual learning, literature-to-skill extraction, self-evolving agent harnesses, and code-grounded verification.

Earth observation. Multi-modal satellite and airborne perception across SAR, optical, infrared, and multispectral imagery; open-source benchmarks and toolkits.