Hancheng Min

Postdoctoral Researcher, Electrical and Systems Engineering, University of Pennsylvania
Email: hanchmin [at] seas [dot] upenn [dot] edu

profile.png

I am a Postdoc Researcher at Center for Innovation in Data Engineering and Science (IDEAS), University of Pennsylvania, advised by Prof. René Vidal. My research centers around building mathematical principles that facilitates the interplay between machine learning and dynamical systems. Recently, I am mainly interested in analyzing gradient-based optimization algorithms on overparametrized neural networks from a dynamical system perspective.

Prior to entering Penn, I received Ph.D. degree from Johns Hopkins University, where I am fortunate to be advised by Prof. Enrique Mallada and Prof. René Vidal. Before Hopkins, I received Master’s degree in Systems Engineering from University of Pennsylvannia and Bachelor’s degree in Automation from Tongji Univerisity, Shanghai.



Recent Update

[Feb, 26, 2025] Our paper Concept Lancet: Decomposing and Transplanting Representations for Diffusion-Based Image Editing is accepted to CVPR 2025 !
[Jan, 22, 2025] Our paper Understanding the Learning Dynamics of LoRA: A Gradient Flow Perspective on Low-Rank Adaptation in Matrix Factorization is accepted to AISTATS 2025 !
[Oct, 25, 2024] Our paper A Frequency Domain Analysis of Slow Coherency in Networked Systems is accepted to Automatica !
[May, 01, 2024] Our paper Can Implicit Bias Imply Adversarial Robustness? is accepted to ICML 2024 !
[Jan, 30, 2024] Our paper Oscillations-Aware Frequency Security Assessment via Efficient Worst-Case Frequency Nadir Computation is accepted to PSCC 2024 !
and more...


Recent publications

  1. colan.png
    Concept Lancet: Decomposing and Transplanting Representations for Diffusion-Based Image Editing
    Jinqi LuoTianjiao DingKwan Ho Ryan ChanHancheng Min, Chris Callison-Burch, and René Vidal
    Conference on Computer Vision and Pattern Recognition (CVPR), 2025 Abs
  2. lora.png
    Understanding the Learning Dynamics of LoRA: A Gradient Flow Perspective on Low-Rank Adaptation in Matrix Factorization
    Ziqing Xu, Hancheng MinJinqi LuoLachlan Ewen MacDonald, Salma Tarmoun, Enrique Mallada, and René Vidal
    International Conference on Artificial Intelligence and Statistics (AISTATS), 2025 Abs PDF
  3. prelu.gif
    Can Implicit Bias Imply Adversarial Robustness?
    Hancheng Min, and René Vidal
    International Conference on Machine Learning (ICML), 21–27 jul 2024 Abs arXiv Bib PDF Poster


Selected publications

  1. coherence.gif
    A Frequency Domain Analysis of Slow Coherency in Networked Systems
    Hancheng MinRichard Pates, and Enrique Mallada
    Automatica, 2025 Abs arXiv Bib
  2. dir_flow.gif
    Early Neuron Alignment in Two-layer ReLU Networks with Small Initialization
    Hancheng MinEnrique Mallada, and René Vidal
    International Conference on Learning Representations (ICLR), May 2024 Abs arXiv Bib PDF Poster Slides
  3. lin_conv.png
    On the Convergence of Gradient Flow on Multi-layer Linear Models
    Hancheng MinRené Vidal, and Enrique Mallada
    International Conference on Machine Learning (ICML), Jul 2023 Abs Bib PDF Poster Slides
  4. inter_area.gif
    Learning Coherent Clusters in Weakly-Connected Network Systems
    Hancheng Min, and Enrique Mallada
    Learning for Dynamics and Control Conference (L4DC), Jun 2023 Abs arXiv Bib PDF Poster
  5. safe_rl.png
    Learning to Act Safely with Limited Exposure and Almost Sure Certainty
    IEEE Transactions on Automatic Control (TAC), May 2023 Abs Bib PDF