Ruihong Zeng (曾锐鸿)

Ruihong Zeng (曾锐鸿)

MBZUAI, Abu Dhabi, UAE

I am a second-year Ph.D. student in Natural Language Processing at MBZUAI, advised by Prof. Preslav Nakov and Prof. Xiuying Chen. I completed my M.S. in Artificial Intelligence at Sun Yat-sen University, advised by Prof. Shangsong Liang.

My research focuses on memory and continual adaptation in LLM agents. I study how agents turn experience into reusable knowledge, coordinate through shared memory, and adapt across long-horizon tasks. I also work on foundation-model evaluation, multimodal fact-checking, and machine unlearning.

Email / Google Scholar / GitHub / CV

Research experience

  • Research Intern · Institute of Foundation Models

    Jul 2026–present

    Supervised by Haonan Li. I refine task specifications and scoring rubrics and develop a capability taxonomy to characterize evaluation coverage. I also investigate context compaction methods that preserve task state and prior decisions across long-horizon workflows.

  • Research Intern · UKP Lab, TU Darmstadt

    Mar–Jun 2026

    Supervised by Prof. Iryna Gurevych. I worked on MIC, a multimodal fact-checking framework for detecting and explaining image–claim inconsistencies using visual evidence and world knowledge.

News

Sep 27, 2026 We released MIC and its code and data.
Sep 24, 2026 RoleMoE has been accepted to NeurIPS 2026! We released RoleMoE and its code and data.
Jan 23, 2025 MGM: Global Understanding of Audience Overlap Graphs for Predicting the Factuality and the Bias of News Media has been accepted to NAACL 2025 as main paper (Oral) !
Dec 16, 2024 We have released the paper On the Structural Memory of LLM Agents.
Dec 12, 2024 We have released the paper MGM: Global Understanding of Audience Overlap Graphs for Predicting the Factuality and the Bias of News Media.
Jul 16, 2024 My first paper Enhancing graph neural networks via memorized global information got accepted in ACM Transactions on the Web.

View all news

Selected publications

* Equal contribution.

2026

  1. NeurIPS
    From Facts to Personas: Interpretable Role Unlearning in LLMs via Mixture-of-Experts
    Ruihong Zeng*, Puning Yang*, Jinghui Zhang, Shen Gao, Xiangliang Zhang, Preslav Nakov, and Xiuying Chen
    In Conference on Neural Information Processing Systems (NeurIPS), 2026
  2. arXiv
    MIC: Explaining Image–Claim Inconsistencies in AI-Generated Multimodal Misinformation
    Ruihong Zeng, Jonathan Tonglet, Preslav Nakov, and Iryna Gurevych
    arXiv preprint, 2026

2025

  1. MGM.png
    MGM: Global Understanding of Audience Overlap Graphs for Predicting the Factuality and the Bias of News Media
    Muhammad Arslan Manzoor*, Ruihong Zeng*, Dilshod Azizov, Preslav Nakov, and Shangsong Liang
    The 2025 Annual Conference of the Nations of the Americas Chapter of the ACL (NAACL), 2025

2024

  1. structural_memory.png
    On the Structural Memory of LLM Agents
    Ruihong Zeng*, Jinyuan Fang*, Siwei Liu, and Zaiqiao Meng
    2024
  2. MGIGNN.png
    Enhancing graph neural networks via memorized global information
    Ruihong Zeng, Jinyuan Fang, Siwei Liu, Zaiqiao Meng, and Shangsong Liang
    ACM Transactions on the Web, 2024