🎉 Our paper "Revisiting Ensembling in One-Shot Federated Learning" has been accepted for publication at NeurIPS 2024.

Our research introduces Fens, a novel approach that bridges the gap between iterative Federated Learning and One-Shot FL. By combining local model training with collaborative aggregator refinement, Fens achieves remarkable results on heterogeneous data distributions. Our experiments with CIFAR-10 demonstrate up to 26.9% higher accuracy than state-of-the-art One-Shot methods while requiring 10.9x less communication than standard Federated Learning. This work represents a significant advancement for privacy-preserving machine learning in bandwidth-limited settings.

The paper will be presented by my co-authors at the conference in Copenhagen in April. Link to the paper.