FedSIR detects noisy-label clients via spectral subspace consistency and relabels their samples using clean-client references, then trains with logit-adjusted loss and distillation to outperform prior FL noise-robust methods on benchmarks.
Compared with CIFAR-10, CIFAR-100 contains a signifi- cantly larger number of classes, which makes the learning problem more challenging under both label noise and non- IID data
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
FedSIR: Spectral Client Identification and Relabeling for Federated Learning with Noisy Labels
FedSIR detects noisy-label clients via spectral subspace consistency and relabels their samples using clean-client references, then trains with logit-adjusted loss and distillation to outperform prior FL noise-robust methods on benchmarks.