🎉 Our paper "Fair Decentralized Learning" has been accepted for publication in the IEEE SaTML 2025.

Facade introduces a novel clustering approach that addresses feature heterogeneity challenges in collaborative training environments. Our method dynamically assigns nodes to appropriate clusters and trains specialized models without requiring prior cluster knowledge. Experimental results show improved model accuracy and fairness with 32.3% lower communication costs on the CIFAR-10 dataset.

The paper will be presented by Sayan at the conference in Copenhagen in April. Link to the paper.