Skin Lesion Classification on HAM10000 (test)
93.32AccuracyFedAvg-SL
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FedAvg-SLLearning Mode=Supervised Learning2022.06 | 93.32 | — | — | 57.09 | 87.58 | 89.64 | 59.09 | — | — | — | — | |
| DiffMICResolution=224x224, Total diffusion time step T=2502023.03 | 90.6 | — | — | 81.6 | — | — | — | — | — | — | — | |
| MAE ViT-H distill EfficientNet-B0 + synthBackbone=EfficientNet-B0, Pre-training=MAE (Teacher), Synthetic Augmentation=Conditional Diffusion, Knowledge Distillation=Distill from ViT-H2026.02 | 90.51 | — | — | 90.1 | — | — | — | — | — | — | — | |
| imFed-Semi2022.06 | 88.94 | — | — | 33.79 | 77.47 | 89.81 | 37.48 | — | — | — | — | |
| ProCoResolution=224x2242023.03 | 88.7 | — | — | 76.3 | — | — | — | — | — | — | — | |
| MAE ViT-H distill to ViT-B (no synth)Backbone=ViT-B, Pre-training=MAE (Teacher), Knowledge Distillation=Distill from ViT-H2026.02 | 87.82 | — | — | 87.97 | — | — | — | — | — | — | — | |
| MAE + ViT-H + synthBackbone=ViT-H, Pre-training=MAE, Synthetic Augmentation=Conditional Diffusion2026.02 | 87.15 | — | — | 87.15 | — | — | — | — | — | — | — | |
| CLResolution=224x2242023.03 | 86.5 | — | — | 73.9 | — | — | — | — | — | — | — | |
| LDAMResolution=224x2242023.03 | 85.7 | — | — | 73.4 | — | — | — | — | — | — | — | |
| CRCKD2021.07 | 85.66 | 76.35 | 78.07 | 76.45 | — | — | — | — | — | — | — | |
| MAE + ViT-L + synthBackbone=ViT-L, Pre-training=MAE, Synthetic Augmentation=Conditional Diffusion2026.02 | 85.12 | — | — | 83.61 | — | — | — | — | — | — | — | |
| Derm-t2im (ViT)Backbone=ViT2026.02 | 85.02 | — | — | — | — | — | — | — | — | — | — | |
| Liu et al.2021.07 | 84.73 | 73.88 | 76.55 | 74.63 | — | — | — | — | — | — | — | |
| FedMatch2022.06 | 84.25 | — | — | 29.13 | 70.9 | 93.33 | 29.56 | — | — | — | — | |
| FedAdam-FM2022.06 | 83.22 | — | — | 27.85 | 70.58 | 92.92 | 28.97 | — | — | — | — | |
| FSSL2022.06 | 83.2 | — | — | 27.9 | 70.86 | 93.39 | 28.32 | — | — | — | — | |
| AdaCBMModel category=Interpretable CBM models, Concept Selection=Concept Utility Selection (ours), k=102024.08 | 82.8 | — | — | — | — | — | — | — | — | — | — | |
| AdaCBMModel category=Interpretable CBM models, Concept Selection=Concept Utility Selection (ours), k=202024.08 | 82.8 | — | — | — | — | — | — | — | — | — | — | |
| MAE ViT-H distill to ViT-B + (non-cond.) synthBackbone=ViT-B, Pre-training=MAE (Teacher), Synthetic Augmentation=Non-conditional Diffusion, Knowledge Distillation=Distill from ViT-H2026.02 | 82.78 | — | — | 78.87 | — | — | — | — | — | — | — | |
| MAE ViT-H distill to ViT-B + synthBackbone=ViT-B, Pre-training=MAE (Teacher), Synthetic Augmentation=Conditional Diffusion, Knowledge Distillation=Distill from ViT-H2026.02 | 82.78 | — | — | 78.87 | — | — | — | — | — | — | — | |
| FedAvg-FM2022.06 | 82.67 | — | — | 29.09 | 70.61 | 91.92 | 30.65 | — | — | — | — | |
| CUFITNoise rate=0.12024.11 | 82.6 | — | — | — | — | — | — | — | — | — | — | |
| DANILResolution=224x2242023.03 | 82.5 | — | — | 67.4 | — | — | — | — | — | — | — | |
| Zhang et al. [29]2021.07 | 82.34 | 72.32 | 74.01 | 72.28 | — | — | — | — | — | — | — | |
| FedProx-FM2022.06 | 82.01 | — | — | 25.21 | 69.86 | 91.45 | 27.87 | — | — | — | — | |
| AdaCBMModel category=Interpretable CBM models, Concept Selection=Concept Utility Selection (ours), k=502024.08 | 81.9 | — | — | — | — | — | — | — | — | — | — | |
| CoDisNoise rate=0.12024.11 | 81.9 | — | — | — | — | — | — | — | — | — | — | |
| OHEMResolution=224x2242023.03 | 81.8 | — | — | 66 | — | — | — | — | — | — | — | |
| Zhang et al. [30]2021.07 | 81.61 | 71.44 | 73.34 | 71.65 | — | — | — | — | — | — | — | |
| EfficientNet-B3Backbone=EfficientNet-B32026.02 | 81.57 | — | — | 73.3 | — | — | — | — | — | — | — | |
| EfficientNet-B7Backbone=EfficientNet-B72026.02 | 81.52 | — | — | 73.61 | — | — | — | — | — | — | — | |
| Co-teachingNoise rate=0.12024.11 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| CUFITNoise rate=0.22024.11 | 81.5 | — | — | — | — | — | — | — | — | — | — | |
| MTLResolution=224x2242023.03 | 81.1 | — | — | 66.7 | — | — | — | — | — | — | — | |
| JoCorNoise rate=0.12024.11 | 81.1 | — | — | — | — | — | — | — | — | — | — | |
| Linear CLSModel category=Non-interpretable models2024.08 | 80.9 | — | — | — | — | — | — | — | — | — | — | |
| Linear-probeModel category=Non-interpretable models2024.08 | 80.7 | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B0Backbone=EfficientNet-B02026.02 | 80.5 | — | — | 75.63 | — | — | — | — | — | — | — | |
| LaBo (10K)Model category=Interpretable CBM models, Concept Selection=Submodular, k=502024.08 | 80.3 | — | — | — | — | — | — | — | — | — | — | |
| CoDisNoise rate=0.22024.11 | 80.1 | — | — | — | — | — | — | — | — | — | — | |
| ViT-B + synthBackbone=ViT-B, Synthetic Augmentation=Conditional Diffusion2026.02 | 79.52 | — | — | 75.06 | — | — | — | — | — | — | — | |
| JoCorNoise rate=0.22024.11 | 79.4 | — | — | — | — | — | — | — | — | — | — | |
| Co-teachingNoise rate=0.22024.11 | 79.1 | — | — | — | — | — | — | — | — | — | — | |
| CUFITNoise rate=0.42024.11 | 79.1 | — | — | — | — | — | — | — | — | — | — | |
| FedIRM2022.06 | 79.05 | — | — | 23.52 | 65.29 | 90.39 | 28.72 | — | — | — | — | |
| Label-free CBMModel category=Interpretable CBM models2024.08 | 78.9 | — | — | — | — | — | — | — | — | — | — | |
| LORA & FTModel category=Non-interpretable models2024.08 | 78.8 | — | — | — | — | — | — | — | — | — | — | |
| ReinNoise rate=0.12024.11 | 78.6 | — | — | — | — | — | — | — | — | — | — | |
| CUFITNoise rate=Mean2024.11 | 78.3 | — | — | — | — | — | — | — | — | — | — | |
| LaBo (10K)Model category=Interpretable CBM models, Concept Selection=Submodular, k=202024.08 | 77.7 | — | — | — | — | — | — | — | — | — | — | |
| Yan et al.2021.07 | 77.16 | 62.71 | 73.37 | 66.7 | — | — | — | — | — | — | — | |
| ViT-L (Baseline)Backbone=ViT-L2026.02 | 76.12 | — | — | 40.43 | — | — | — | — | — | — | — | |
| LaBo (10K)Model category=Interpretable CBM models, Concept Selection=Submodular, k=102024.08 | 75.6 | — | — | — | — | — | — | — | — | — | — | |
| Linear probingNoise rate=0.12024.11 | 75.6 | — | — | — | — | — | — | — | — | — | — | |
| Co-teachingNoise rate=Mean2024.11 | 75.5 | — | — | — | — | — | — | — | — | — | — | |
| CoDisNoise rate=Mean2024.11 | 75.5 | — | — | — | — | — | — | — | — | — | — | |
| JoCorNoise rate=Mean2024.11 | 75.4 | — | — | — | — | — | — | — | — | — | — | |
| Linear probingNoise rate=0.22024.11 | 75.3 | — | — | — | — | — | — | — | — | — | — | |
| LaBoModel category=Interpretable CBM models, Concept Selection=Submodular, k=502024.08 | 74.9 | — | — | — | — | — | — | — | — | — | — | |
| Co-teachingNoise rate=0.42024.11 | 74.3 | — | — | — | — | — | — | — | — | — | — | |
| LaBoModel category=Interpretable CBM models, Concept Selection=Submodular, k=202024.08 | 74.1 | — | — | — | — | — | — | — | — | — | — | |
| CoDisNoise rate=0.42024.11 | 74.1 | — | — | — | — | — | — | — | — | — | — | |
| JoCorNoise rate=0.42024.11 | 73.9 | — | — | — | — | — | — | — | — | — | — | |
| Derm-t2im (MobileNet)Backbone=MobileNet2026.02 | 73.58 | — | — | — | — | — | — | — | — | — | — | |
| LaBoModel category=Interpretable CBM models, Concept Selection=Submodular, k=102024.08 | 73 | — | — | — | — | — | — | — | — | — | — | |
| ViT-B (Baseline)Backbone=ViT-B2026.02 | 72.25 | — | — | 32.09 | — | — | — | — | — | — | — | |
| ReinNoise rate=0.22024.11 | 72.1 | — | — | — | — | — | — | — | — | — | — | |
| Linear probingNoise rate=0.42024.11 | 71 | — | — | — | — | — | — | — | — | — | — | |
| Linear probingNoise rate=Mean2024.11 | 71 | — | — | — | — | — | — | — | — | — | — | |
| CUFITNoise rate=0.62024.11 | 70.1 | — | — | — | — | — | — | — | — | — | — | |
| Backbone FTModel category=Non-interpretable models2024.08 | 67.9 | — | — | — | — | — | — | — | — | — | — | |
| Co-teachingNoise rate=0.62024.11 | 67.3 | — | — | — | — | — | — | — | — | — | — | |
| JoCorNoise rate=0.62024.11 | 67.1 | — | — | — | — | — | — | — | — | — | — | |
| Full-trainingNoise rate=0.12024.11 | 66.5 | — | — | — | — | — | — | — | — | — | — | |
| CoDisNoise rate=0.62024.11 | 66.1 | — | — | — | — | — | — | — | — | — | — | |
| Full-trainingNoise rate=0.22024.11 | 62.6 | — | — | — | — | — | — | — | — | — | — | |
| Linear probingNoise rate=0.62024.11 | 61.9 | — | — | — | — | — | — | — | — | — | — | |
| Full-trainingNoise rate=Mean2024.11 | 61.3 | — | — | — | — | — | — | — | — | — | — | |
| ReinNoise rate=Mean2024.11 | 60.8 | — | — | — | — | — | — | — | — | — | — | |
| Full-trainingNoise rate=0.62024.11 | 59.9 | — | — | — | — | — | — | — | — | — | — | |
| Full-trainingNoise rate=0.42024.11 | 56.1 | — | — | — | — | — | — | — | — | — | — | |
| ReinNoise rate=0.42024.11 | 54.9 | — | — | — | — | — | — | — | — | — | — | |
| ReinNoise rate=0.62024.11 | 37.8 | — | — | — | — | — | — | — | — | — | — |