Sebastian U. Stich

Sebastian U. Stich is a tenured faculty member at the CISPA Helmholtz Center for Information Security, where he has been since December 2021. He is also a member of the European Lab for Learning and Intelligent Systems (ELLIS) since June 2020. His research focuses on optimization methods for machine learning, collaborative learning algorithms, privacy and security in distributed systems, and theoretical foundations of deep learning.

Stich received his PhD in Theoretical Computer Science from ETH Zurich (2014), following a Master's in Mathematics at the same institution (2010-2014). Prior to CISPA, he worked as a research scientist at EPFL (2016-2021) and held positions at ETH Zurich and ICTEAM/CORE. He has been awarded the ERC Consolidator Grant 2024, Google Research Scholar Award (2023), and Meta Privacy-Enhancing Technologies Research Award (2022).

His team includes Dr. Anton Rodomanov (since 2023), Dr. Rotem Mulayoff (since 2024), Xiaowen Jiang (2023), Yuan Gao (2023), and notable alumni like Anastasia Koloskova (defended 2023). Stich actively organizes workshops (e.g., NeurIPS OPT 2024) and serves on editorial boards (Journal of Optimization Theory and Applications, Transactions on Machine Learning Research). He teaches advanced courses in optimization at Saarland University and has held visiting positions at MIT.

Key scientific contributions include:

• Developing ProgFed for progressive federated learning (2021)
• Creating ProxSkip to accelerate communication in federated settings (2022)
• Formalizing SCAFFOLD with control variates for FL (2020)
• Introducing RelaySum mechanism for decentralized learning (2021)
• Proposing Lookahead-Minmax for GAN training (2021)

His work addresses fundamental challenges in:

• Decentralized optimization theory
• Communication-efficient algorithms
• Privacy-preserving model training
• Handling heterogeneous data distributions
• Stochastic gradient dynamics
• Second-order optimization methods