Trust based update screening boosts federated learning with label attacks
Peer-Relative Trust-Guided Update Screening for Robust Semi-Decentralized Federated Learning under Label-Flipping Attacks
Cryptography and SecurityDistributed, Parallel, and Cluster ComputingNeural and Evolutionary Computing
Summary
Federated learning lets many computers train a model together without sharing raw data, but some participants might send bad info to mess it up. The authors improve a system that trusts participants based on their history and compares their updates to peers, helping to spot dishonest or harmful contributions. This new way keeps most good updates and blocks many bad ones, especially when devices’ data is unevenly spread out. Their method helps make collaborative learning more reliable even when some participants try to sabotage it.
What this means in practice
- •For federated learning engineers: Improve robustness of federated training systems against label-flipping attacks by screening client updates using peer-relative trust and update quality metrics.
- •For blockchain developers: Integrate off-chain trust assessments into blockchain-based audit and incentive mechanisms for semi-decentralized federated learning.
Authors
Ajay Kumar Shrestha, Sandhya Joshi
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.