Alex Bui

The paper identifies a new class of difficult-to-detect trustworthiness failures, termed 'Silent Failures,' that arise when personalizing foundation models using federated learning, arguing that current evaluation methods are insufficient.