HeteRo-Select: Informativeness as the Participation Driver in Heterogeneous Federated Learning
Researchers have developed a new framework called HeteRo-Select for federated learning systems that prioritizes data informativeness over link speed for gradient compression. This approach aims to address the issue where bandwidth and data informativeness can become misaligned in non-IID data scenarios. By using an informativeness score to guide client selection, compression ratios, and server aggregation weights, HeteRo-Select has demonstrated significant improvements in speed and reductions in traffic, even outperforming bandwidth-driven methods when these signals are deliberately anti-correlated. AI
IMPACT Optimizes federated learning efficiency by prioritizing data informativeness over bandwidth, potentially leading to faster training and reduced resource usage.