Image Classification on MNIST i.i.d. (test)
97.053Test AccuracyFEDADAVR
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FEDADAVROptimiser=ADAM, Server learning rate (ηs)=0.012026.01 | 97.053 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADAM, Server learning rate (ηs)=0.012026.01 | 96.906 | — | — | — | — | — | |
| FEDADAVROptimiser=YOGI, Server learning rate (ηs)=0.012026.01 | 96.657 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADABELIEF, Server learning rate (ηs)=0.012026.01 | 96.651 | — | — | — | — | — | |
| FEDADAVROptimiser=ADABELIEF, Server learning rate (ηs)=0.0052026.01 | 96.535 | — | — | — | — | — | |
| FEDADAVROptimiser=ADABELIEF, Server learning rate (ηs)=0.012026.01 | 96.472 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=YOGI, Server learning rate (ηs)=0.0052026.01 | 96.286 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADABELIEF, Server learning rate (ηs)=0.0052026.01 | 96.28 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=YOGI, Server learning rate (ηs)=0.012026.01 | 96.023 | — | — | — | — | — | |
| FEDADAVROptimiser=ADAM, Server learning rate (ηs)=0.0052026.01 | 95.827 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADAM, Server learning rate (ηs)=0.0052026.01 | 95.793 | — | — | — | — | — | |
| FEDADAVROptimiser=YOGI, Server learning rate (ηs)=0.0052026.01 | 95.368 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADAGRAD, Server learning rate (ηs)=0.012026.01 | 94.542 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADABELIEF, Server learning rate (ηs)=0.0012026.01 | 93.656 | — | — | — | — | — | |
| FEDADAVROptimiser=ADABELIEF, Server learning rate (ηs)=0.0012026.01 | 93.106 | — | — | — | — | — | |
| FEDADAVROptimiser=ADAGRAD, Server learning rate (ηs)=0.012026.01 | 92.781 | — | — | — | — | — | |
| DFedReweightingT=0.12025.12 | 92.12 | — | — | — | — | 0.85 | |
| DFedReweightingT=0.52025.12 | 91.996 | — | — | — | — | 0.838 | |
| q-FDFLq=0.012025.12 | 91.873 | — | — | — | — | 0.787 | |
| DFedAvg2025.12 | 91.867 | — | — | — | — | 0.807 | |
| DFedReweightingT=0.012025.12 | 91.846 | — | — | — | — | 0.896 | |
| FEDADAVR-QUANTOptimiser=ADAGRAD, Server learning rate (ηs)=0.0052026.01 | 91.78 | — | — | — | — | — | |
| q-FDFLq=0.12025.12 | 91.76 | — | — | — | — | 0.964 | |
| FEDADAVR-QUANTOptimiser=YOGI, Server learning rate (ηs)=0.0012026.01 | 91.659 | — | — | — | — | — | |
| FEDADAVROptimiser=ADAGRAD, Server learning rate (ηs)=0.0052026.01 | 91.595 | — | — | — | — | — | |
| q-FDFLq=0.52025.12 | 91.254 | — | — | — | — | 0.608 | |
| FedVARPServer learning rate (ηs)=1.02026.01 | 90.604 | — | — | — | — | — | |
| FEDADAVROptimiser=LAMB, Server learning rate (ηs)=0.012026.01 | 90.108 | — | — | — | — | — | |
| FEDADAVROptimiser=ADAM, Server learning rate (ηs)=0.0012026.01 | 89.946 | — | — | — | — | — | |
| FEDADAVROptimiser=YOGI, Server learning rate (ηs)=0.0012026.01 | 89.739 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADAM, Server learning rate (ηs)=0.0012026.01 | 89.497 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=LAMB, Server learning rate (ηs)=0.012026.01 | 88.685 | — | — | — | — | — | |
| FEDADAVROptimiser=ADAGRAD, Server learning rate (ηs)=0.0012026.01 | 80.725 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=ADAGRAD, Server learning rate (ηs)=0.0012026.01 | 80.639 | — | — | — | — | — | |
| FEDADAVR-QUANTOptimiser=LAMB, Server learning rate (ηs)=0.0052026.01 | 78.681 | — | — | — | — | — | |
| FedMPDDModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, m=400, projection ratio (d)=2%2025.12 | 77.37 | 0.052 | — | — | — | — | |
| FEDADAVROptimiser=LAMB, Server learning rate (ηs)=0.0052026.01 | 77.128 | — | — | — | — | — | |
| lp-projModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=602025.12 | 73.01 | 0.069 | — | 0.75 | — | — | |
| FedMPDDModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, m=600, projection ratio (d)=3%2025.12 | 67.75 | 0.079 | — | — | — | — | |
| Top-kModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, k=4002025.12 | 65.75 | 0.077 | — | 0.89 | — | — | |
| FedMPDDBytes Budget=2000000, Target Acc=60, m=2002025.12 | 65.29 | — | — | 0.03 | 1,536,000 | — | |
| FedMPDDModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, m=800, projection ratio (d)=4%2025.12 | 58.49 | 0.093 | — | — | — | — | |
| FedMPDDBytes Budget=2000000, Target Acc=60, m=4002025.12 | 57.62 | — | — | 0.14 | 2,432,000 | — | |
| FedMPDDBytes Budget=2000000, Target Acc=60, m=6002025.12 | 48.85 | — | — | 0.13 | 3,456,000 | — | |
| QSGDBytes Budget=2000000, Target Acc=60, bit-width=42025.12 | 38.92 | — | — | 0.88 | 5,343,440 | — | |
| FEDADAVROptimiser=LAMB, Server learning rate (ηs)=0.0012026.01 | 38.891 | — | — | — | — | — | |
| FedVARPServer learning rate (ηs)=0.12026.01 | 33.715 | — | — | — | — | — | |
| QSGDBytes Budget=2000000, Target Acc=60, bit-width=82025.12 | 28.86 | — | — | 0.99 | 10,681,440 | — | |
| FEDADAVR-QUANTOptimiser=LAMB, Server learning rate (ηs)=0.0012026.01 | 21.756 | — | — | — | — | — | |
| QSGDModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, Quantization=8-bit2025.12 | 21.66 | 0.376 | — | 0.98 | — | — | |
| FedVARPServer learning rate (ηs)=0.012026.01 | 11.948 | — | — | — | — | — | |
| FedSGDModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=602025.12 | 11.45 | 1.439 | — | 1 | — | — | |
| FedSGD + LaplaceModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, variance=12025.12 | 11.41 | 1.869 | — | — | — | — | |
| FedSGD + LaplaceModel=LeNet, Bytes Budget (GB)=0.09, Target Acc (%)=60, variance=0.52025.12 | 11.13 | 1.611 | — | — | — | — | |
| FedSGDBytes Budget=2000000, Target Acc=60, Noise=None2025.12 | — | — | — | 1 | 40,192,000 | — | |
| FedSGDBytes Budget=2000000, Target Acc=60, Noise=Gaussian, Variance=0.12025.12 | — | — | — | 0.8 | 42,704,000 | — | |
| FedSGDBytes Budget=2000000, Target Acc=60, Noise=Gaussian, Variance=12025.12 | — | — | — | 0.59 | 45,216,000 | — | |
| FedSGDBytes Budget=2000000, Target Acc=60, Noise=Laplace, Variance=0.12025.12 | — | — | — | 0.82 | 42,704,000 | — | |
| FedSGDBytes Budget=2000000, Target Acc=60, Noise=Laplace, Variance=12025.12 | — | — | — | 0.6 | 45,216,000 | — |