I am a fourth-year PhD student at Tsinghua University, advised by Prof. Chun Yuan and Li Shen. My research centers on data-efficient model reuse (i.e., learning from models): how to transfer, compose, and extend the knowledge encoded in pretrained models while reducing reliance on task-specific data and costly training from scratch.

My approach combines empirical investigation with theoretical analysis: I identify recurring phenomena, formulate and test hypotheses, and develop principled methods informed by these insights. A central goal is to establish effective and scalable approaches to model reuse for vision models, large language models (LLMs), and multimodal large language models (MLLMs), with careful attention to data efficiency and computational cost.

πŸ“ Selected Publications

My work explores data-efficient model reuse from three angles:

01 Model Merging

Closed-Form Spectral Regularization for Multi-Task Model Merging β€” research overview

Closed-Form Spectral Regularization for Multi-Task Model Merging

Yongxian Wei, Runxi Cheng, Xingxuan Zhang, Li Shen, Chun Yuan, Peng Cui, Dacheng Tao

Preprint

OptMerge: Unifying Multimodal LLM Capabilities and Modalities via Model Merging β€” research overview

OptMerge: Unifying Multimodal LLM Capabilities and Modalities via Model Merging

Yongxian Wei, Runxi Cheng, Weike Jin, Enneng Yang, Li Shen, Lu Hou, Sinan Du, Chun Yuan, Xiaochun Cao, Dacheng Tao

ICLR 2026

Adaptive gradient projection preserves shared knowledge while merging task experts

Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent

Yongxian Wei, Anke Tang, Li Shen, Zixuan Hu, Chun Yuan, Xiaochun Cao

ICML 2025

Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectors β€” research overview

Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectors

Runxi Cheng*, Feng Xiong*, Yongxian Wei, Wanyun Zhu, Chun Yuan

ICML 2025

02 Synthetic Data

LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence β€” research overview

LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence

Xingxuan Zhang, Gang Ren, …, Yongxian Wei (Core Contributor), …, Peng Cui

Technical Report

Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis β€” research overview

Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis

Yongxian Wei, Yilin Zhao, Li Shen, Xinrui Chen, Runxi Cheng, Sinan Du, Hao Yu, Chun Yuan, Dian Li

Preprint

Diffusion generates synthetic images, which form tasks for meta-learning

Meta-Learning without Data via Unconditional Diffusion Models

Yongxian Wei, Zixuan Hu, Li Shen, Zhenyi Wang, Chun Yuan

IEEE TCSVT

03 Model Inversion

Open-Vocabulary Customization from CLIP via Data-Free Knowledge Distillation β€” research overview

Open-Vocabulary Customization from CLIP via Data-Free Knowledge Distillation

Yongxian Wei, Zixuan Hu, Li Shen, Zhenyi Wang, Chun Yuan, Dacheng Tao

ICLR 2025 (Oral)

Task-Distributionally Robust Data-Free Meta-Learning β€” research overview

Task-Distributionally Robust Data-Free Meta-Learning

Zixuan Hu, Yongxian Wei, Li Shen, Zhenyi Wang, Baoyuan Wu, Chun Yuan, Dacheng Tao

IEEE TPAMI

Unlocking Tuning-Free Few-Shot Adaptability in Visual Foundation Models by Recycling Pre-Tuned LoRAs β€” research overview

Unlocking Tuning-Free Few-Shot Adaptability in Visual Foundation Models by Recycling Pre-Tuned LoRAs

Zixuan Hu, Yongxian Wei, Li Shen, Chun Yuan, Dacheng Tao

CVPR 2025

Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models β€” research overview

Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models

Yongxian Wei, Zixuan Hu, Li Shen, Zhenyi Wang, Chun Yuan, Dacheng Tao

ICML 2024

Sparse Model Inversion: Efficient Inversion of Vision Transformers for Data-Free Applications β€” research overview

Sparse Model Inversion: Efficient Inversion of Vision Transformers for Data-Free Applications

Zixuan Hu, Yongxian Wei, Li Shen, Zhenyi Wang, Chun Yuan, Dacheng Tao

ICML 2024

FREE: Faster and Better Data-Free Meta-Learning β€” research overview

FREE: Faster and Better Data-Free Meta-Learning

Yongxian Wei, Zixuan Hu, Zhenyi Wang, Li Shen, Chun Yuan, Dacheng Tao

CVPR 2024

πŸ”₯ Honors

  • National Scholarship for Graduate Students (for the second time), 2025.10
  • National Scholarship for Graduate Students (at my first year), 2024.10
  • Undergraduate President’s Medal (Top 10 university-wide), 2022.10
  • National Scholarship for Undergraduate Students, 2021.10
  • The First Runner-Up (2/116) of ICCV LargeFineFoodAI, 2021.10

πŸ“– Education

  • PhD student, Tsinghua University, 2023 – Present
  • Undergrad, Nanjing University of Science and Technology, 2019 – 2023

πŸ’Ό Academic Service

  • Reviewer (Conferences): ICML / ICLR / NeurIPS / CVPR / ICCV / ECCV / AAAI
  • Reviewer (Journals): TPAMI / TIP / TKDE