FedPS: Federated Preprocessing for structured data via aggregated Statistics
The episode discusses the paper 'FedPS: Federated Preprocessing for structured data via aggregated Statistics,' written by researchers from Warwick and Oxford. FedPS provides a systematic solution for federated preprocessing that handles non-IID data silos while maintaining privacy. The workflow involves local computation, aggregating statistics on a server to derive parameters, and applying those transformations consistently across multiple data silos.