Get Started
Home
Topics
Search
Library
Research questionHow do topology and peer aggregation affect semantic backdoor persistence in decentralized federated learning?Semantic triggers can be less conspicuous than synthetic patches, but their effects may vary across samples. Repeated peer aggregation in decentralized networks can either dilute or propagate these backdoor signals depending on the network structure and attacker presence.
AI
Alignment & Safety
Machine Learning
Statistical Machine Learning
Technology
Latest papersRecent research connected to this question, newest first.CACTUS: Mask-Guided Semantic Clean-Label Backdoors in Decentralized Federated LearningThe evidence covers decentralized federated learning across speech, text, tabular, and image tasks, using nine aggregation rules. It examines topology and malicious-node-ratio sensitivity, including experiments with 30% malicious nodes, and reports attack success rates rather than deployment defenses.research paper · Sep 2, 2026
Related questions
How can federated learning protect client updates from inference while resisting Byzantine manipulation?How can collaborative learning preserve graph structure when graph data are distributed across agents?How can federated learning reject poisoned client updates without server validation data or participant history?How can adversarial perturbations transfer to unseen semantic-segmentation models while accounting for dense spatial and class-wise structure?