Your Neighbors Know: Leveraging Local Neighborhoods for Backdoor Detection in Decentralized Learning

Decentralized learning (DL) is an emerging machine learning paradigm where nodes collaboratively train models without a central server. However, the collaborative nature of DL makes it vulnerable to backdoor attacks, where a model is taught to behave normally on standard inputs while executing hidde...

Descripció completa

Dades bibliogràfiques

Autors principals:	Biswas, Sayan, Boutet, Antoine, Frey, Davide, Gaudel, Romaric, Guerraoui, Rachid, Jacovella, Maxime, Kermarrec, Anne-Marie, Lerévérend, Dimitri, Taïani, François, de Vos, Martijn
Format:	Article
Publicat:	2026
Matèries:	Machine Learning
Accés en línia:	https://arxiv.org/abs/2605.19969 http://arxiv.org/abs/2605.19969
Etiquetes:	Afegir etiqueta Sense etiquetes, Sigues el primer a etiquetar aquest registre!