Zixi (Charlie) Chen

prof_pic.jpg

I am a Computer Science PhD student at NYU Courant, advised by Prof. Andrew Gordon Wilson. My research broadly focuses on building more reliable and aligned intelligent systems, with specific interests in scaling laws, novel architectures/optimizers, and structured matrices.

Previously, I interned at Q Labs, working on improving scaling laws through architecture interventions, and at Amazon SCOT, working on causal inference for time series. Prior to my PhD, I completed a BA/MS accelerated program at NYU, earning my Bachelor’s in Honors Mathematics and Computer Science and a Master’s in Mathematics.

selected publications

  1. Figure 1: transformer architectures comparing boundary operators, shared weights, and model growth
    How Model Growth, Recursion, and Boundary Operators Influence Scaling Exponents
    Zixi Chen, Akshay Vegesna, Samip Dahal, and Andrew Gordon Wilson
    arXiv preprint arXiv:2609.19107, 2026
  2. Figure 1: random emergence times across model sizes and attention patterns before and after a capability emerges
    Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns
    Vatsal Baherwani, Zixi Chen, Shikai Qiu, Andrew Gordon Wilson, and Pavel Izmailov
    arXiv preprint arXiv:2606.25010, 2026
  3. A Unifying Approach to Self-Organizing Systems Interacting via Conservation Laws
    A Unifying Approach to Self-Organizing Systems Interacting via Conservation Laws
    Frank Barrows, Guanming Zhang, Satyam Anand, Zixi Chen, Jonathan Lin, Aman Desai, Stefano Martiniani, and Francesco Caravelli
    Advanced Intelligent Discovery, 2026
  4. Hyperparameter Transfer Enables Consistent Gains of Matrix-Preconditioned Optimizers Across Scales
    Hyperparameter Transfer Enables Consistent Gains of Matrix-Preconditioned Optimizers Across Scales
    Shikai Qiu*, Zixi Chen*, Hoang Phan, Qi Lei, and Andrew Gordon Wilson
    In Advances in Neural Information Processing Systems (NeurIPS), 2025
  5. Searching for Efficient Linear Layers over a Continuous Space of Structured Matrices
    Searching for Efficient Linear Layers over a Continuous Space of Structured Matrices
    Andres Potapczynski*, Shikai Qiu*, Marc Finzi, Christopher Ferri, Zixi Chen, Micah Goldblum, Bayan Bruss, Christopher De Sa, and Andrew Gordon Wilson
    In Advances in Neural Information Processing Systems (NeurIPS), 2024