Jannis Halbey

prof_pic.jpg

I am a fourth-year PhD candidate at the Institure of Mathematics, TU Berlin under the supervision of Sebastian Pokutta. I am also a member of the IOL group at Zuse Institute Berlin and the Berlin Mathematical School (BMS).

My research interests lie in large-scale optimization, convex geometry, and combinatorial optimization, with a particular focus on projection-free methods such as Frank–Wolfe algorithms and their variants. I am interested in the development and analysis of scalable algorithms for convex, constrained optimization, especially correction and active-set strategies. More recently, I have worked on lower bounds and convergence analysis for first-order methods, aiming to better understand fundamental limits of these algorithms. Furthermore, I am actively contributing to the FrankWolfe.jl package.

news

Oct 01, 2026 Supervising again the seminar on Modern topics in machine learnign and optimization by Sebastian Pokutta at TU Berlin.
Sep 25, 2026 Our paper Curvature-Dependent Lower Bounds for Frank-Wolfe got accepted at NeurIPS 2026 and I was recongnized as a “Top-Reviewer” of the conference (top 8%).
Sep 24, 2026 Gave a talk on “Curvature-Dependent Lower Bounds for Frank-Wolfe” at GHOST Inria in Grenoble.
May 01, 2026 Our paper Lower Bounds for Frank-Wolfe on Strongly Convex Sets got accepted at ICML 2026.
Apr 03, 2026 Supervising the seminar on Modern topics in machine learnign and optimization by Sebastian Pokutta at TU Berlin.

selected publications

  1. lp-lower.png
    Curvature-Dependent Lower Bounds for Frank-Wolfe
    Jannis Halbey, Christophe Roux, and Sebastian Pokutta
    arXiv preprint arXiv:2605.10595, 2026
  2. l2-lower.png
    Lower Bounds for Frank-Wolfe on Strongly Convex Sets
    Jannis Halbey, Daniel Deza, Max Zimmer, and 3 more authors
    arXiv preprint arXiv:2602.04378, 2026
  3. quadcorrections.png
    Efficient quadratic corrections for Frank-Wolfe algorithms
    Jannis Halbey, Seta Rakotomandimby, Mathieu Besançon, and 2 more authors
    Advances in Neural Information Processing Systems, 2026