Fig. 1: FedOcw for Parkinson’s disease detection from speech across five scenarios, showing data distributions and model performance.
From: FedOcw: optimized federated learning for cross-lingual speech-based Parkinson’s disease detection
Different combinations of multilingual datasets assigned to clients: (Scenarios A) Spanish–Italian, (Scenarios B) Spanish–Chinese, (Scenarios C) Italian–Chinese, (Scenarios D) Spanish–Italian–Czech, and (Scenarios E) Spanish–Italian–Chinese–Czech–English. Each sub-panel shows data allocation across clients (e.g., C0–C3: Spanish, C4–C7: Italian, etc.) and box plots comparing the performance of five federated learning methods—FedAvg, FedProx, Scaffold, FedNova, and the proposed FedOcw—on client test data. Box plots indicate performance distributions, where the center line marks the median, the circle denotes the mean, box limits correspond to the 1st and 3rd quartiles, whiskers span 1.5 times the interquartile range, and outliers are shown individually.
