Prof. Dr. Sebastian Pokutta
Vice President and Division Head
Mathematical Algorithmic Intelligence
AI in Society, Science, and Technology (AIS²T)
Zuse Institute Berlin (ZIB)
Professor for
Optimization and Machine Learning
Institute of Mathematics
Electrical Engineering and Computer Science (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]
(Informal) TL;DR. We use computers to learn from data and make better decisions.
Recent Papers.
- Kerdreux, T., Roux, C., d’Aspremont, A., and Pokutta, S. (2021). Linear Bandits on Uniformly Convex Sets. Preprint. [arXiv] [summary]
- Chmiela, A., Khalil, E., Gleixner, A., and Pokutta, S. (2021). Learning to Schedule Heuristics in Branch-and-Bound. Preprint. [arXiv]
- Carderera, A., Diakonikolas, J., Lin, C. Y., and Pokutta, S. (2021). Parameter-free Locally Accelerated Conditional Gradients. Preprint. [arXiv]
- Kerdreux, T., d’Aspremont, A., and Pokutta, S. (2021). Local and Global Uniform Convexity Conditions. Preprint. [arXiv]
- Braun, G., and Pokutta, S. (2021). Dual Prices for Frank-Wolfe Algorithms. Preprint. [arXiv]
- Combettes, C. W., and Pokutta, S. (2021). Complexity of Linear Minimization and Projection on Some Sets. Preprint. [arXiv] [code]
- Pokutta, S. (2021). Mathematik, Machine Learning und Artificial Intelligence. To Appear in Mitteilungen Der DMV (German). [PDF]
- Carderera, A., Pokutta, S., Schütte, C., and Weiser, M. (2021). CINDy: Conditional gradient-based Identification of Non-linear Dynamics – Noise-robust recovery. Preprint. [arXiv] [summary]
- Kerdreux, T., d’Aspremont, A., and Pokutta, S. (2021). Projection-Free Optimization on Uniformly Convex Sets. To Appear in Proceedings of AISTATS. [arXiv] [summary]
- Combettes, C. W., Spiegel, C., and Pokutta, S. (2020). Projection-Free Adaptive Gradients for Large-Scale Optimization. Preprint. [arXiv] [summary] [code]
- Pokutta, S., Spiegel, C., and Zimmer, M. (2020). Deep Neural Network Training with Frank-Wolfe. Preprint. [arXiv] [summary] [code]
- Carderera, A., and Pokutta, S. (2020). Second-order Conditional Gradient Sliding. Preprint. [arXiv] [summary] [code]
Select Recent Talks and Teaching.
- 02/2021: (technical) “Structured ML Training via Conditional Gradients”. Talk at IPAM Deep Learning and Combinatorial Optimization Workshop (online). [slides] [video]
- 11/2020: (technical) “Conditional Gradients: Overview and Recent Advances”. Talk at RWTH Aachen Mathematical Colloquium (online). [slides]
- 11/2020: (general) “AI and-for-with-against Humanity?”. Keynote at Human-centric Artificial Intelligence: 2nd French-German-Japanese Symposium (online). [slides]
- 09/2020: (technical) “Restarting Algorithms: Sometimes there is Free Lunch”. Keynote at CPAIOR 2020 (online). [slides] [video]
- 09/2020: (technical) “Robust ML Training with Conditional Gradients”. Talk at CO@Work 2020 Summer School (online). [slides] [video]
- SoSe/2021: Discrete Optimization and Machine Learning (seminar)
Recent Blog Posts.
- 04/2021: Linear Bandits on Uniformly Convex Sets
- 01/2021: CINDy: Conditional gradient-based Identification of Non-linear Dynamics
- 11/2020: DNN Training with Frank–Wolfe
- 10/2020: Projection-Free Adaptive Gradients for Large-Scale Optimization
- 09/2020: Accelerating Domain Propagation via GPUs
News.
- 11/2020: Our group received a Google Research Award to support our work on Integer Programming solvers.
- 10/2020: Four projects funded by the Math+ Research Center.
- 10/2020: HTW Berlin enters cooperation with Zuse Institute Berlin. ZIB Press Release, HTW Press Release (German)
- 05/2020: Our Special Priority Program (SPP) proposal ‘Theoretical Foundations of Deep Learning’ was funded by DFG. With an overall budget of EUR 8.5m, this program sets out to significantly boost our fundamental understanding of Deep Learning. The coordination team is Gitta Kutyniok (speaker), Martin Burger, Matthias Hein, Sebastian Pokutta, and Ingo Steinwart. DFG Press Release (German), TUB Press Release (German) SPP Homepage
- 04/2020: Research Campus MODAL enters second funding phase