From October 13th to 19th, 2024, Professor Lúcia Drummond from Fluminense Federal University (UFF) in Brazil visited INRIA Bordeaux as part of the DecoHPC associate team collaboration. Drummond, a faculty member in UFF’s Department of Computer Science since 1989, is a renowned expert in high-performance computing (HPC) and cloud computing. She also leads a major project that connects her research group, C+HPC Lab at UFF, with Petrobras, the Brazilian oil company.
During her time at INRIA, Drummond worked closely with Laercio Pilla, a researcher from the Storm/Topal team, and Alan Lira, a PhD student co-supervised by both. Their collaborative efforts focused on advancing federated learning systems, which enable distributed machine learning across multiple clients, each using private local data. This approach is crucial in scenarios where sharing raw data is not possible due to privacy concerns.
The team explored ways to optimize client selection in federated learning, an essential task given the resource and data diversity across clients. Their work introduced two innovative algorithms—MEC and ECMTC—that balance both time and energy consumption, ensuring a more efficient and effective global model.
