Hao-Zhe Tan @ LAMDA, NJU-SZ

tanhz.jpg 

谈昊哲
Hao-Zhe Tan
M.Sc. Student, LAMDA Group
School of Intelligence Science and Technology
Nanjing University (Suzhou Campus), Suzhou 215163, China

Supervisor: Lan-Zhe Guo (郭兰哲)
Laboratory: Nanyong Building, Suzhou Campus of Nanjing University
E-mail: tanhz@lamda.nju.edu.cn

Biography

Currently, I am a graduate student of School of Intelligence Science and Technology in Nanjing University and a member of LAMDA Group, which is led by professor Zhi-Hua Zhou.

I received my B.Sc. degree from School of Computer Science in June 2023 from Nanjing University of Information Science & Technology. In the same year, I was admitted to study for a M.Sc. degree in Nanjing University.

Research Interests

My research interests include Machine Learning and Data Mining. Most recently, I am interested in Learnware and Multimodal Reasoning.

Publications - Preprints

WSFG

  • Hao-Zhe Tan, Zhi Zhou, Lan-Zhe Guo, Yu-Feng Li. Pre-Trained Vision-Language Model Selection and Reuse for Downstream Tasks. arXiv:2501.18271. [Paper]

    • We propose a novel paradigm called Model Label Learning, which includes the processes of model labeling, selection, and reuse. This paradigm is both time- and data-efficient, and highly scalable. It can give birth to new VLM model hubs, which can simplifying user selection and reuse of VLMs for their tasks.

WSFG

  • Zhi Zhou*, Hao-Zhe Tan*, Peng-Xiao Song, Lan-Zhe Guo. CGI: Identifying Conditional Generative Models with Example Images. arXiv:2501.13991. [Paper]

    • We present a novel systematic solution, namely, Prompt-Based Model Indetification (PMI). Specifically, we first introduce Automatic Specification Assignment to generate specifications using a pre-defined prompt set or developer-provided prompt set. Then, Requirement Generation abstracts the user requirements. Both specification and requirement are projected into a unified model matching space for future model identification. Finally, we propose a Task-Specific Matching mechanism to adjust specification according to the user requirements in the model matching space to precisely identify the most suitable model.

(* denotes equal contribution)

Awards & Honors

Service

Correspondence