Yuping Zheng

Ph.D. Student in Computer Science · University of Virginia

I am a Ph.D. student in Computer Science at the University of Virginia, advised by Prof. Xi Peng.

My research interests broadly lie in representation learning, generative modeling, and interpretability. Recently, I have been particularly interested in understanding how models learn and organize internal representations, especially in world models and multimodal systems.

Email · CV · Google Scholar · GitHub
Yuping Zheng

News

Aug. 2026 I joined UVA Computer Science as a Ph.D. student, advised by Prof. Xi Peng.
Jun. 2026 Our paper Vaxjo 2.0 was published in Frontiers in Cellular and Infection Microbiology.
Dec. 2025 Our preprint VaxjoGNN was released on bioRxiv.
2025 Our paper Efficiently Democratizing Medical LLMs for 50 Languages via a Mixture of Language Family Experts was accepted to ICLR 2025.

Publications

2026

Overview of the Vaxjo 2.0 project workflow

Vaxjo 2.0: An Ontology- and Large Language Model-Powered Knowledge Base of Vaccine Adjuvants and Mechanisms

Joshua Monickaraj, Hasin Rehana, Ani Bernardi, Yuping Zheng, Taiyu Lin, Le Liu, Amogh Madireddi, Leo Yeh, Jie Zheng, Junguk Hur, Yongqun He

Frontiers in Cellular and Infection Microbiology, 2026

Paper

2025

Overview of the VaxjoGNN disease-adjuvant ranking pipeline

VaxjoGNN: A Graph Neural Network for Ontology-Grounded Vaccine Adjuvant Recommendation

Yuping Zheng, Yongqun He

bioRxiv, 2025

An ontology-grounded graph learning framework for vaccine adjuvant recommendation under sparse and heterogeneous biomedical evidence.

Paper

Hybrid-k routing in Apollo-MoE

Efficiently Democratizing Medical LLMs for 50 Languages via a Mixture of Language Family Experts

Guorui Zheng, Xidong Wang, Juhao Liang, Nuo Chen, Yuping Zheng, Benyou Wang

International Conference on Learning Representations (ICLR), 2025

A multilingual mixture-of-experts framework for scaling medical language models across 50 languages through language-family-based routing.

Paper · Code

2024

Overview of the VaxLLM design and workflow

VaxLLM: Leveraging Fine-tuned Large Language Model for Automated Annotation of Brucella Vaccines

Xingxian Li, Yuping Zheng, Joy Hu, Jie Zheng, Zhigang Wang, Yongqun He

bioRxiv, 2024

Paper · Code

Research

NFDM project figure

Learning Forward Processes for Discrete Flow Matching

Can the forward process of a discrete generative model be learned rather than hand-designed? We explore learnable corruption schedules and discrete state representations for flow matching. A learned one-dimensional codebook improves generation, while increasing its dimensionality reveals an unexpected failure mode: the training objective continues to improve even as the learned representation becomes highly anisotropic and generation quality deteriorates.

VaxjoGNN project figure

Ontology-Grounded Learning for Vaccine Adjuvant Recommendation

How can structured biomedical knowledge help learning when experimental evidence is sparse and heterogeneous? In VaxjoGNN, we use an ontology-grounded graph representation to integrate vaccine, adjuvant, and biological knowledge for vaccine adjuvant recommendation.

Paper

Education

2026–Present

University of Virginia

Ph.D. in Computer Science

Advisor: Prof. Xi Peng

2022–2026

The Chinese University of Hong Kong, Shenzhen

B.S. in Data Science and Big Data Technology