Newest papers

Peer-Relative Trust-Guided Update Screening for Robust Semi-Decentralized Federated Learning under Label-Flipping Attacks

Ajay Kumar Shrestha, Sandhya Joshi · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.25714 · cited by 0 works (OpenAlex) · open-access PDF

German

Vertrauens- und Reputations-Kernmechanismen können den Einfluss bösartiger Kunden im federierten Lernen (FL) verringern, aber das strenge Update-Filtern ist unter nicht-unabhängigen und identisch verteilten (nicht-IID) Daten schwierig, weil ehrliche lokale Updates die Leistung auf einem geteilten Validierungs-Proxy vorübergehend verringern können. Aufbauend auf unserem früheren vertrauensbasierten semi-decentralisierten Influencer FL (SDFL) -Framework operationalisiert dieses Papier die Vertrauens-Studie-Signale und führt ein Paar-Relative-Update-Screening ein, das eine robuste Validierung-Utility-Score mit Richtungs- und Norm-basierter Update-Qualität kombin

Machine translation (Meta M2M100); the English original follows.

Abstract

Trust and reputation mechanisms can reduce the influence of malicious clients in federated learning (FL), but strict update filtering is difficult under non-independent and identically distributed (non-IID) data because honest local updates may temporarily reduce performance on a shared validation proxy. Building on our earlier trust-based semi-decentralized FL (SDFL) framework, this paper operationalizes the trust signals and introduces peer-relative update screening that combines a robust validation-utility score with direction- and norm-based update quality. The method is evaluated with 50 MNIST clients, 20% persistent label-flip attackers, Dirichlet heterogeneity levels of 0.5 and 0.1, 100 communication rounds, and five paired random seeds. Under moderate heterogeneity, the proposed method reaches 98.37% final accuracy while rejecting 95.82% of malicious updates and retaining 90.30% of benign updates; the final benign and malicious trust distributions are completely separated (KS=1.00). Under severe heterogeneity, it reaches 95.25% accuracy, rejects 88.55% of malicious updates, and retains 76.90% of benign updates (KS=0.96). Trust weighting alone preserves slightly higher utility in the most heterogeneous condition, but accepts all malicious submis-sions. The results therefore distinguish influence mitigation from explicit containment and quantify the utility-selectivity trade-off introduced by screening. The study provides an empirically validated off-chain governance core that can feed blockchain-based audit and incentive layers in SDFL.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Weighting, Reputation, Federated learning, Independent and identically distributed random variables, Audit

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange