Federated Image Classification on MNIST non-IID, α=0.1 (test)
95.51Test AccuracyFedAvg-M
Evaluation results
| Method | Links | |
|---|---|---|
| FedAvg-MCompression ratio=1, Communication rounds (T)=100, Dirichlet concentration (α)=0.1, Number of clients (p)=50, Local epochs per round=1, Mini-batch size (b)=32, Global step-size (η)=0.0182026.04 | 95.51 | |
| FedSLoPCompression ratio=1/7, Communication rounds (T)=100, Dirichlet concentration (α)=0.1, Number of clients (p)=50, Local epochs per round=1, Mini-batch size (b)=32, Global step-size (η)=0.018, FedSLoP projection rank r=112, FedSLoP momentum=0.82026.04 | 95.11 | |
| FedLoRA-MCompression ratio=1/7, Communication rounds (T)=100, Dirichlet concentration (α)=0.1, Number of clients (p)=50, Local epochs per round=1, Mini-batch size (b)=32, Global step-size (η)=0.018, LoRA rank=152026.04 | 92.47 | |
| FedMefCompression ratio=1/7, Communication rounds (T)=100, Dirichlet concentration (α)=0.1, Number of clients (p)=50, Local epochs per round=1, Mini-batch size (b)=32, Global step-size (η)=0.0182026.04 | 89.65 | |
| NeuLiteCompression ratio=1/7, Communication rounds (T)=100, Dirichlet concentration (α)=0.1, Number of clients (p)=50, Local epochs per round=1, Mini-batch size (b)=32, Global step-size (η)=0.0182026.04 | 89.23 | |
| FederatedSelectCompression ratio=1/7, Communication rounds (T)=100, Dirichlet concentration (α)=0.1, Number of clients (p)=50, Local epochs per round=1, Mini-batch size (b)=32, Global step-size (η)=0.0182026.04 | 87.59 |