Image Classification on CIFAR-10 ρ=0.8 (test)
71.45AccuracyHARMONY
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HARMONYBackbone=VGG-11, Number of Clients (N)=50, Data Partitioning (Shard)=22026.05 | 71.45 | 89.58 | |
| SplitGPBackbone=VGG-11, Number of Clients (N)=50, Data Partitioning (Shard)=2, Personalization Intensity (λ)=0.22026.05 | 67.71 | 50.28 | |
| SplitGP + FTBackbone=VGG-11, Number of Clients (N)=50, Data Partitioning (Shard)=2, Personalization Intensity (λ)=0.2, Evaluation Protocol (Fine-tuning)=on-device fine-tuning2026.05 | 64.46 | 94.9 | |
| Multi-exit SFLBackbone=VGG-11, Number of Clients (N)=50, Data Partitioning (Shard)=22026.05 | 61.01 | 66.54 | |
| SplitGPBackbone=VGG-11, Number of Clients (N)=50, Data Partitioning (Shard)=2, Personalization Intensity (λ)=0.92026.05 | 54 | 86.75 | |
| SplitFedBackbone=VGG-11, Number of Clients (N)=50, Data Partitioning (Shard)=22026.05 | 45.03 | 188.63 |