FedSumUp Reduces Privacy Leakage in Federated Learning via Lossy Compression

FedSumUp implements a learnable lossy compressor on client devices to generate compact, visually interpretable representations for federated intelligence. This approach mitigates reconstruction attacks while allowing users to audit the privacy-utility tradeoff without high client-side computational overhead.