Zhenhan Huang

I am a

Hi there! My name is Zhenhan Huang (黄臻瀚). I am a third year graduate student in the department of computer science, Rensselaer Polytechnic Institute, Troy, New York, United States. I am fortunate enough to be advised by Prof. Jianxi Gao.

My research interests align broadly with deep learning in computer vision and natural language processing. My current research focuses on meta learning and multimodal deep learning.

News

  • [2026.02] Our work "Unveil Fundamental Graph Properties for Neural Architecture Search" was accepted by Advanced Science.
  • [2026.01] I did 2026 spring research intern at Vector Institute. My research topic is the robustness of multi-agent system in medical application.
  • [2025.05] I did 2025 summer research intern at Vector Institute. My research topic is multi-agent collaboration.
  • [2025.04] Our work "Differentiable Prompt Learning for Vision Language Models" was accepted by IJCAI 2025.
  • [2025.04] Our work "Language Models Are Good Tabular Learners" was accepted by TMLR.
  • [2024.12] Our work "Modular Prompt Learning Improves Vision-Language Models" was accepted by ICASSP 2025.
  • [2024.07] Our work "Graph is all you need? Lightweight Data-Agnostic Neural Architecture Search Without Training" was accepted by AutoML 2024 workshop.
  • [2024.04] Our work "Network Properties Determine Neural Network Performance" was accepted by Nature Communication.
  • [2023.10] I was fortunate to be supported by IBM-RPI Future of Computing Research Collaboration (FCRC) grant.
  • [2023.05] I went to IBM Thomas J. Watson Research Center for 2023 summer intern. My research topic was to adapt large language models in the tabular data domain.
  • [2022.08] I was enrolled in the Ph.D. program in computer science of Rensselaer Polytechnic Institute.

Education

  • Ph.D. in Computer Science, Rensselaer Polytechnic Institute
    2022 - Present, GPA: 4.0/4.0
  • M.S. in Computer Science, Rensselaer Polytechnic Institute
    2021 - 2022, GPA: 4.0/4.0
  • Ph.D. in Materials Engineering, Rensselaer Polytechnic Institute
    2017 - 2022, GPA: 3.8/4.0
  • M.S. in Materials Engineering, Harbin Institute of Technology
    2015 - 2017, GPA: 3.5/4.0
  • B.S. in Materials Engineering, Harbin Institute of Technology
    2011 - 2015, GPA: 3.7/4.0

Publication

  1. Chunheng Jiang*, Zhenhan Huang*, Tejaswini Pedapati, Pin-Yu Chen, Yizhou Sun and Jianxi Gao. Network Properties Determine Neural Network Performance. Nature Communication. (* Equal contribution). [PDF][Code]
  2. Zhenhan Huang, Tejaswini Pedapati, Pin-Yu Chen, Chunheng Jiang and Jianxi Gao. Graph is All You Need? Lightweight Data-Agnostic Neural Architecture Search Without Training. AutoML 2024 Workshop. [PDF] [Code]
  3. Aamod Khatiwada, Harsha Kokel, Ibrahim Abdelaziz, Subhajit Chaudhury, Julian Dolby, Oktie Hassanzadeh, Zhenhan Huang, Tejaswini Pedapati, Horst Samulowitz, Kavitha Srinivas. TabSketchFM: Sketch-based Tabular Representation Learning for Data Discovery over Data Lakes. NeurIPS 2024 Workshop. ICDE 2025. [PDF]
  4. Zhenhan Huang, Tejaswini Pedapati, Pin-Yu Chen and Jianxi Gao. Modular Prompt Learning Improves Vision-Language Models. ICASSP 2025. [PDF][Code]
  5. Zhenhan Huang, Kavitha Srinivas, Horst Samulowitz, Niharika S. D'Souza, Charu C. Aggarwal, Pin-Yu Chen, Jianxi Gao. Language Models Are Good Tabular Learners. TMLR. [PDF][Code]
  6. Zhenhan Huang, Tejaswini Pedapati, Pin-Yu Chen and Jianxi Gao. Differentiable Prompt Learning for Vision Language Models. IJCAI 2025. [PDF][Code][Project]
  7. Zhenhan Huang, Tejaswini Pedapati, Pin-Yu Chen, Chunheng Jiang and Jianxi Gao. Unveil Fundamental Graph Properties for Neural Architecture Search. Advanced Science. [PDF][Code][Project]

Research Directions

Neural Architecture Search

We aimed to bridge neural architecture space to graph space. The connection was based on training dynamics or forward propagation. Neural architectures were evaluated through the lens of graphs. Related work:

  • Network Properties Determine Neural Network Performance
  • Unveil Fundamental Graph Properties for Neural Architecture Search

Natural Language Processing

We aimed to adapt transformer-based language model to tabular data domain. Embedding approaches were proposed to help language model to understand structured data. Unsupervised pretraining and supervised fine-tuning were applied in the training scheme. Related work:

  • TabSketchFM: Sketch-based Tabular Representation Learning for Data Discovery over Data Lakes
  • Language Models Are Good Tabular Learners

Parameter-Efficient Fine-Tuning

We aimed to introduce heterogeneous design in the prompting methods for vision-language models. Through prompting, pretrained foundational vision-language models were adapted to downstream tasks. Related work:

  • Modular Prompt Learning Improves Vision-Language Models
  • Differentiable Prompt Learning for Vision Language Models

Contact Info

Feel free to reach out with any questions or potential collaborations: