Prof. Dr. Sebastian Pokutta

Vice President
Zuse Institute Berlin (ZIB)

Professor for
Optimization and Machine Learning
Mathematics and EECS (courtesy)
Technische Universität Berlin

---

Research Lab. My group is interested in Artificial Intelligence, Optimization, and Machine Learning. We develop new methodologies (e.g., new optimization and learning algorithms), work on combining learning and decision-making, as well as design AI Systems for real-world deployment in various application contexts. [more about research and projects]

(Informal) TL;DR. We use computers to learn from data and make better decisions.

Prospective Students. If you are interested in working in our group or writing your MS/BS thesis please only use the email applications-aisst@zib.de.

Recent Papers.

1. Designolle, S., Iommazzo, G., Besançon, M., Knebel, S., Gelß, P., and Pokutta, S. (2023+). Improved local models and new Bell inequalities via Frank-Wolfe algorithms. To Appear in Physical Reviews Research. [arXiv] [slides]
2. Thuerck, D., Sofranac, B., Pfetsch, M., and Pokutta, S. (2023). Learning Cuts via Enumeration Oracles. To Appear in Proceedings of NeurIPS. [arXiv]
3. Zimmer, M., Spiegel, C., and Pokutta, S. (2023). Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging. Preprint. [arXiv]
4. Deza, A., Onn, S., Pokutta, S., and Pournin, L. (2023). Kissing polytopes. Preprint. [arXiv]
5. Stengl, M., Gelß, P., Klus, S., and Pokutta, S. (2023). Existence and Uniqueness of Solutions of the Koopman–von Neumann Equation on Bounded Domains. Preprint. [arXiv]
6. Martinez-Rubio, D., Wirth, E., and Pokutta, S. (2023). Accelerated and Sparse Algorithms for Approximate Personalized PageRank and Beyond. Proceedings of COLT. [arXiv] [slides] [poster]
7. Martinez-Rubio, D., Roux, C., Criscitiello, C., and Pokutta, S. (2023). Accelerated Methods for Riemannian Min-Max Optimization Ensuring Bounded Geometric Penalties. Preprint. [arXiv]
8. Martinez-Rubio, D., and Pokutta, S. (2023). Accelerated Riemannian Optimization: Handling Constraints to Bound Geometric Penalties. Proceedings of COLT. [arXiv] [poster]
9. Aigner, K., Bärmann, A., Braun, K., Liers, F., Pokutta, S., Schneider, O., Sharma, K., and Tschuppik, S. (2023). Data-driven Distributionally Robust Optimization over Time. To Appear in INFORMS Journal on Optimization. [arXiv]
10. Kreimeier, T., Pokutta, S., Walther, A., and Woodstock, Z. (2023). On a Frank-Wolfe Approach for Abs-smooth Functions. Preprint. [arXiv]
11. Wirth, E., Kera, H., and Pokutta, S. (2023). Approximate Vanishing Ideal Computations at Scale. Proceedings of ICLR. [arXiv] [slides] [poster]
12. Braun, G., Pokutta, S., and Weismantel, R. (2022). Alternating Linear Minimization: Revisiting von Neumann’s alternating projections. Preprint. [arXiv] [slides] [video]
13. Braun, G., Carderera, A., Combettes, C. W., Hassani, H., Karbasi, A., Mokthari, A., and Pokutta, S. (2022). Conditional Gradient Methods. Preprint. [arXiv]

Select Recent Talks and Teaching.

• 09/2023: (technical) “The approximate Carathéodory problem and an application to quantum mechanics”. Plenary at 7th Workshop on Future Algorithms and Applications (Berlin, Germany).
• 08/2023: (technical) “Alternating Linear Minimization: Revisiting von Neumann’s alternating projections”. Talk at ICIAM 2023 (Tokyo, Japan). [slides]
• 08/2023: (technical) “Improved local models and new Bell inequalities via Frank-Wolfe algorithms”. Talk at DO x ML conference (Tokyo, Japan). [slides]
• 07/2023: (technical) “Alternating Linear Minimization: Revisiting von Neumann’s alternating projections”. Talk at Jon-Shmuel Halfway to Twelfty (Paris, France). [slides]
• 05/2023: (technical) “Alternating Linear Minimization: Revisiting von Neumann’s alternating projections”. Talk at University of Magdeburg MathCoRe Lecture (Magdeburg, Germany). [slides]
• WS/2022: Discrete Optimization and Machine Learning (seminar)

Recent Blog Posts.

• 08/2023: Accelerated and Sparse Algorithms for Approximate Personalized PageRank and Beyond
• 08/2023: Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging
• 07/2023: How I Learned to Stop Worrying and Love Retraining
• 07/2023: Alternating Linear Minimization
• 05/2023: Improved local models and new Bell inequalities via Frank-Wolfe algorithms

News.

• 05/2023: Received Gödel Prize together with Samuel Fiorini, Serge Massar, Hans Raj Tiwary, Ronald de Wolf, and Thomas Rothvoss
• 05/2023: We are organizing the fifth conference on “Discrete Optimization and Machine Learning” in Aug 2023 at GRIPS in Tokyo.
• 02/2023: We are organizing a Thematic Einstein Semester on “Mathematical Optimization for Machine Learning” within the Math+ Cluster of Excellence. The semester consists of various activities throughout the semester with three workshops, a conference, and a summer school as some of the highlights. We are looking forward to seeing you in Berlin!
• 11/2022: We finished our monograph on Frank-Wolfe methods a.k.a. Conditional Gradients. [arxiv] [webpage] [blog]
• 06/2022: Symposium on Theory of Computing (STOC) Test of Time award (10 years) for “Linear vs. Semidefinite Extended Formulations: Exponential Separation and Strong Lower Bounds”, S. Fiorini, S. Massar, S. Pokutta, H.R. Tiwary, R. de Wolf from 2012.