Revealing the influence of participant failures on model quality in cross-silo Federated Learning

Researchers conducted extensive experiments to analyze how participant failures affect Federated Learning model quality across different data types and scenarios. The study reveals that data skewness significantly impacts model performance and can lead to overly optimistic evaluations when participants are missing from the training process.