Zhenyi Wang

Postdoctoral Associate at University of Maryland, College Park

prof_pic.jpg

Brendan Iribe Center for Computer Science and Engineering

8125 Paint Branch Drive

College Park, Maryland

I’m Zhenyi Wang, currently a Postdoctoral Associate at University of Maryland, College Park, working with Prof. Heng Huang. Previously, I received my PhD from Department of Computer Science and Engineering at University at Buffalo in 2023, working with Prof. Mingchen Gao.

Research Summary: My long-term research goal is to develop artificial general intelligence (AGI) that is data-efficient, capable of continual learning, responsible, secure, and able to reuse and assemble knowledge. Specifically, I have developed algorithms across the following research areas: (1) Continual learning. (2) AI safety for responsible AGI, focusing on protecting model intellectual property and ensuring appropriate usage through applicability authorization mechanisms. (3) Knowledge reuse and reassembly from existing pre-trained models. (4) Data-efficient learning.

Specifically:

  • Continual Learning: (1) Memorization and Generalization Trade-Off [ICML2022], (2) Generalized Distributionally Robust Memory Evolution [TPAMI2023], (3) Low-Rank Memory Parameterization [NeurIPS2023], (4) Remember and Forget Trade-Off [ICLR2024], (5) Old Task and New Task Trade-Off [NeurIPS2024]
  • AI Safety, Deep Model Intellectual Property Protection: (1) Applicability/Usage Authorization [ICLR2024], (2) Efficient Model Extraction Defense [NeurIPS2023, ICML2024]
  • Model Merge and Reuse: (1) Adaptive Model Merging [ICLR2024], (2) Representation Surgery [ICML2024], (3) Model Grouping [ICML2024]
  • Data Efficient Learning: (1) Bayesian Meta Sampling for Few-Shot Learning Uncertainty Adaptation [ICLR2020], (2) Neural Tangent Kernel for Inner-Loop-Free Few-Shot Learning [ICLR2021], (3) Few-Shot Learning in Evolving Domains [ICCV2021, CVPR2022]

Please refer to Google Scholar for more papers

News

  • November 2024 One paper has been accepted by TPAMI.

  • September 2024 One paper has been accepted by NeurIPS 2024.

  • September 2024 One paper has been accepted by TPAMI.

  • July 2024 Two papers accepted at ECCV 2024.

  • May 2024 Four papers accepted at ICML 2024.

  • February 2024 One paper accepted at CVPR 2024.

  • January 2024 Three papers accepted at ICLR 2024.

  • September 2023 Two papers has been accepted by NeurIPS 2023.

  • September 2023 One paper has been accepted by TPAMI.