Research Area

Distributed Machine Learning

The MSRG members working on distributed deep learning focus on three topics: performance optimization, privacy-preserving and federated learning systems, and graph processing.

Our performance optimization efforts range from designing efficient data preprocessing pipelines over cost-optimal geo-distributed training of large models to brining foundation models to embedded hardware for privacy preserving training.
Our federated learning efforts center around energy-optimal utilization of modern AI accelerators for billion-parameter models. We also do research at the intersection of legal and tech for federated learning to set the pathway for future research priorities.

A special application area for distributed learning in our group are Graph Neural Networks. As we work on GNN systems, we are interested in optimizing the resource utilization for graph (pre-)processing and GNN training.