Get Started
Home
Topics
Search
Library
Research questionHow can federated learning reject poisoned client updates without server validation data or participant history?Malicious clients can submit updates that manipulate or degrade the global model. The server may lack trusted validation data and reliable prior knowledge about participating clients.
AI
Alignment & Safety
Machine Learning
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.FedReview: Review and Dispose Poisoned Updates without Validation Datasets or Historic KnowledgeThe source describes a federated-learning setting where randomly selected clients review other updates using their own training data. Reviewers rank updates, and the server uses majority voting to identify suspicious updates for removal; experiments examine adversarial environments.research paper · Sep 1, 2026
Related questions
How can federated learning protect client updates from inference while resisting Byzantine manipulation?How can federated learning stay reliable when heterogeneous edge clients are intermittently available and availability changes over time?How can federated unlearning prevent classifier broadcasts from revealing and enabling reinsertion of deleted client data?How can federated learning protect confidence calibration from attacks that preserve accuracy?