Florian Hübler

Doctoral student, Department of Computer Science, ETH Zurich

About

I am a doctoral student in computer science at ETH Zurich, working with Niao He in the Optimization and Decision Intelligence Group. Previously, I completed an MSc in Data Science at ETH Zurich and a BSc in Mathematics at the Technical University of Munich.

I am looking for an industry internship in summer 2027.

Research

My research focuses on optimisation for machine learning, combining practical optimiser design with convergence theory and complexity lower bounds. My current focus is understanding and designing matrix-aware optimisers such as Muon and Muown for pre-training large language models. I am also interested in optimisation under heavy-tailed noise and parameter-agnostic (parameter-free) methods for non-convex problems.

Publications and Preprints (Google Scholar)

  1. The Local Geometry of Magnitude and Direction Decoupling
    C. Zhang, K. Lion, F. Hübler, J. Pang, A. Orvieto
    OPT 2026 workshop at NeurIPS. OpenReview
  2. Muown Implicitly Performs Angular Step-size Decay
    F. Hübler*, K. Lion*, A. Orvieto, N. He
    OPT 2026 workshop at NeurIPS (Oral). arXiv · code
  3. Muown: Row-Norm Control for Muon Optimization
    K. Lion, F. Hübler, B. Li, A. Orvieto, N. He
    Preprint, 2026. arXiv · code
  4. Free Heavy-Tailed Lunch for Muon: A Theoretical Justification of Empirical Success
    F. Hübler, T. Pethick, S. Sra
    NeurIPS 2026 (Spotlight, top 1.3% of submissions). arXiv · code
  5. Can SGD Handle Heavy-Tailed Noise?
    I. Fatkhullin, F. Hübler, G. Lan
    SIAM Journal on Optimization, to appear. arXiv
  6. From Gradient Clipping to Normalization for Heavy Tailed SGD
    F. Hübler*, I. Fatkhullin*, N. He
    AISTATS 2025. PMLR 258:2413–2421. PMLR
  7. Parameter-Agnostic Optimization under Relaxed Smoothness
    F. Hübler, J. Yang, X. Li, N. He
    AISTATS 2024. PMLR 238:4861–4869. PMLR · code

*Equal contribution.

Talks

Teaching

Supervision

Service

Contact

Email
florian.huebler (at) inf.ethz.ch
GitHub
github.com/fhueb
Google Scholar
profile
LinkedIn
profile