About Me

Welcome! I am a newly graduated PhD seeking Postdoctoral positions. My research lies at the fascinating intersection of Statistical Physics and Deep Learning Theory.

I completed my PhD at the School of Physics, Sun Yat-sen University under the supervision of Prof. Haiping Huang. Prior to my doctoral studies, I obtained both my M.S. in Theoretical Physics (under the supervision of Prof. Quanhui Liu) and my B.S. in Applied Physics from Hunan University. My doctoral research focused on developing mathematical frameworks to understand the loss landscapes, optimization dynamics, and generalization capabilities of deep neural networks using tools from statistical mechanics (e.g., spin glass theory, replica methods, and dynamical mean-field theory).

I am highly motivated to collaborate with researchers in both physics and computer science departments to unravel the theoretical mysteries of artificial intelligence.

Job Market Status: I am actively seeking Postdoctoral research opportunities beginning in Jun 2026. Please feel free to reach out via email!

News

  • May 2026
    🎉 Successfully defended my Ph.D. dissertation titled "High-Dimensional Statistical Physics of Learning in Neural Networks"!
  • Dec 2025
    Our paper on using response function to measure consciousness was published in Physical Review Research.
  • Nov 2025
    Our paper on synaptic plasticity and chaos transition in neural networks was published in Physical Review E.
  • Sep 2025
    Preprint on the geometric origin of adversarial vulnerability is now uploaded to arXiv.

Research Interests

Loss Landscapes & Generalization

Analyzing the geometric properties of high-dimensional loss surfaces of deep neural networks using spin glass theory and replica methods.

Gradient Descent Dynamics

Studying the non-equilibrium dynamical behavior of stochastic gradient descent (SGD) and its role in implicit regularization.

Infinite-Width Limits

Investigating Neural Tangent Kernels (NTK) and mean-field limits of neural networks from a statistical mechanics perspective.

Generative Models

Exploring the theoretical foundations of diffusion models and autoregressive transformers using dynamical system theories.

Selected Publications

  • Synaptic plasticity alters the nature of the chaos transition in neural networks
    Wenkang Du and Haiping Huang
    Physical Review E, 112(5): 054208, 2025
    Journal Link
  • Response function as a quantitative measure of consciousness in brain dynamics
    Wenkang Du and Haiping Huang
    Physical Review Research, 7(4): 043249, 2025
    Journal Link
  • Geometric origin of adversarial vulnerability in deep learning
    Y. Ren, W. K. Du, J. Zhou, et al.
    arXiv preprint arXiv:2509.01235, 2025 (Under Review)
    arXiv Link

Contact

If you are interested in my research or have potential postdoc openings, please don't hesitate to get in touch!

Email: dwklan1905@gmail.com