The group of Sebastian Stich works on the design and analysis of optimization methods for machine learning, with research topics ranging from decentralized and federated optimization, adaptive stochastic algorithms, and generalization in deep learning, to robustness, privacy, and security aspects of modern ML systems.

Current research directions include the ERC Consolidator Grant CollectiveMinds, which aims to make machine learning more sustainable and efficient by enabling multiple smaller models to collaborate, adapt, and forget outdated knowledge intelligently, and a DFG-funded project on distributed optimization and scalable training of deep neural networks, including transformer architectures.

PhD students can work on any of the topics spanned by the group, or on closely related areas that meaningfully connect to our research. We expect candidates to explain their research interests and their connection to the group’s activities in their cover letter.

Possible research topics include: Collaborative, federated or decentralized learning/Continual learning and knowledge sharing/Optimization algorithms for large-scale or distributed training/Robustness, fairness, and personalization in multi-agent learning/Training efficiency and communication reduction/Distributed training of transformer models