I am a doctoral researcher at Zuse Institute Berlin (ZIB) in the Interactive Optimization and Learning (IOL) research group and a PhD candidate at the Institute of Mathematics, TU Berlin under the supervision of Sebastian Pokutta. I am also a member of the Berlin Mathematical School (BMS).

My research interests lie at the intersection of (continuous) optimization and machine learning. In particular, I am interested in optimization on Riemannian manifolds, online optimization and federated learning.

E-mail: roux@zib.de

Publications

* means equal contribution

1. From Associations to Activations: Comparing Behavioral and Hidden-State Semantic Geometry in LLMs [arXiv]
   L. Schiekiera, M. Zimmer, C. Roux, S. Pokutta, F. Günther
2. SparseSwaps: Tractable LLM Pruning Mask Refinement at Scale [arXiv]
   M. Zimmer, C. Roux, M. Wagner, D. Hendrych, S. Pokutta
3. A Free Lunch in LLM Compression: Revisiting Retraining after Pruning [arXiv]
   M. Wagner, C. Roux, M. Zimmer, S. Pokutta
4. Don't Be Greedy, Just Relax! Pruning LLMs via Frank-Wolfe [arXiv]
   C. Roux*, M. Zimmer*, A. d’Aspremont, S. Pokutta
5. Implicit Riemannian Optimism with Applications to Min-Max Problems [ICML 2025] [arXiv]
   C. Roux*, D. Martínez-Rubio*, S. Pokutta
6. Accelerated Methods for Riemannian Min-Max Optimization Ensuring Bounded Geometric Penalties [AISTATS 2025] [arXiv]
   D. Martínez-Rubio*, C. Roux*, C. Criscitiello, S. Pokutta
7. On the Byzantine-Resilience of Distillation-Based Federated Learning [ICLR 2025] [arXiv] [summary]
   C. Roux*, M. Zimmer*, S. Pokutta
8. Convergence and Trade-Offs in Riemannian Gradient Descent and Riemannian Proximal Point [ICML 2024] [arXiv]
   D. Martínez-Rubio*, C. Roux*, S. Pokutta
9. Efficient Online-Bandit Strategies for Minimax Learning Problems [arXiv]
   C. Roux, E. Wirth, S. Pokutta, T. Kerdreux
10. Linear Bandits on Uniformly Convex Sets [JMLR] [arXiv] [summary]
    T. Kerdreux, C. Roux, A. d’Aspremont, S. Pokutta