FedDDL improves federated out-of-distribution generalization by generating background-mixed counterfactual samples and aligning clients with causal prototypes, yielding an average Top-1 gain of about 4.5 percent over nine baselines.
A cross-client coordinator in fed- erated learning framework for conquering heterogeneity
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Federated Deconfounding and Debiasing Learning for Out-of-Distribution Generalization
FedDDL improves federated out-of-distribution generalization by generating background-mixed counterfactual samples and aligning clients with causal prototypes, yielding an average Top-1 gain of about 4.5 percent over nine baselines.