Image Classification on CIFAR-100 (Accuracy)
91.2AccuracyAdapterTune
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| AdapterTuneBackbone=ViT-B/16, Trainable Parameters (%)=0.9%2026.03 | 91.2 | — | — | — | |
| Procedural warm-upBackbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=Procedural warm-up (ours), Evaluation Protocol=Fine-tuned2025.11 | 89.2 | — | — | — | |
| Mimetic init.Backbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=Mimetic init. [51], Evaluation Protocol=Fine-tuned2025.11 | 88.78 | — | — | — | |
| FractalDBBackbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=FractalDB [23] warm-up, Evaluation Protocol=Fine-tuned2025.11 | 88.35 | — | — | — | |
| Random init.Backbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=Random init. [10], Evaluation Protocol=Fine-tuned2025.11 | 87.54 | — | — | — | |
| iBOTBackbone=ViT-L/16, Pre-training Epochs=250, Training Heuristics=w/ heuristics, Linear Probe Training Epochs=102026.06 | 87.2 | — | — | — | |
| Linear ProbeBackbone=CLIP ViT-B/162025.10 | 86.4 | — | — | — | |
| iBOTBackbone=ViT-B/16, Pre-training Epochs=400, Training Heuristics=w/ heuristics, Linear Probe Training Epochs=102026.06 | 85.9 | — | — | — | |
| I-JEPABackbone=ViT-H/14, Pre-training Epochs=300, Training Heuristics=w/ heuristics, Linear Probe Training Epochs=102026.06 | 85.5 | — | — | — | |
| SEG-MIL-CBMBackbone=CLIP ViT-B/162025.10 | 85.26 | — | — | — | |
| MoCoV3Backbone=ViT-B/16, Pre-training Epochs=300, Training Heuristics=w/ heuristics, Linear Probe Training Epochs=102026.06 | 85.2 | — | — | — | |
| DINOBackbone=ViT-B/16, Pre-training Epochs=400, Training Heuristics=w/ heuristics, Linear Probe Training Epochs=102026.06 | 85 | — | — | — | |
| AdapterTuneBackbone=ViT-S/16, Trainable Parameters (%)=0.9%2026.03 | 84.9 | — | — | — | |
| TaDABackbone=ViT-L/16, Number of samples=5002026.06 | 84.8 | — | — | — | |
| Task onlyBackbone=ViT-L/16, Number of samples=5002026.06 | 84.6 | — | — | — | |
| VISRegBackbone=ViT-L/14, Pre-training Epochs=100, Training Heuristics=w/o heuristics, Linear Probe Training Epochs=1002026.06 | 84.2 | — | — | — | |
| DARE-TIESBackbone=ViT-L/16, Number of samples=5002026.06 | 84.1 | — | — | — | |
| data2vecBackbone=ViT-L/14, Pre-training Epochs=1600, Training Heuristics=w/ heuristics, Linear Probe Training Epochs=102026.06 | 83.7 | — | — | — | |
| LeJEPA*Backbone=ViT-L/14, Pre-training Epochs=100, Training Heuristics=w/o heuristics, Linear Probe Training Epochs=1002026.06 | 83.7 | — | — | — | |
| SVD MergeBackbone=ViT-L/16, Number of samples=5002026.06 | 83.7 | — | — | — | |
| Mag. PruningBackbone=ViT-L/16, Number of samples=5002026.06 | 83.4 | — | — | — | |
| LinearBackbone=ViT-L/16, Number of samples=5002026.06 | 83 | — | — | — | |
| TIESBackbone=ViT-L/16, Number of samples=5002026.06 | 83 | — | — | — | |
| DARE-LinBackbone=ViT-L/16, Number of samples=5002026.06 | 83 | — | — | — | |
| DN-CBMBackbone=CLIP ViT-B/162025.10 | 82.1 | — | — | — | |
| Base modelBackbone=ViT-L/16, Number of samples=5002026.06 | 82 | — | — | — | |
| Domain onlyBackbone=ViT-L/16, Number of samples=5002026.06 | 81.8 | — | — | — | |
| Head-only tuningBackbone=ViT-B/16, Trainable Parameters (%)=0.1%2026.03 | 81.5 | — | — | — | |
| LaBoBackbone=CLIP ViT-B/162025.10 | 81.2 | — | — | — | |
| Transformer (ANN)T=12026.03 | 81.02 | — | — | — | |
| Full fine-tuningBackbone=ViT-B/16, Trainable Parameters (%)=100%2026.03 | 80.7 | — | — | — | |
| CDMBackbone=CLIP ViT-B/162025.10 | 80.5 | — | — | — | |
| AdapterTuneBackbone=DeiT-T, Trainable Parameters (%)=0.9%2026.03 | 80.3 | — | — | — | |
| MorphANNT=12026.03 | 80.29 | — | — | — | |
| Task Arith.Backbone=ViT-L/16, Number of samples=5002026.06 | 80 | — | — | — | |
| MorphSNNT=42026.03 | 79.87 | — | — | — | |
| Full fine-tuningBackbone=DeiT-T, Trainable Parameters (%)=100%2026.03 | 79.7 | — | — | — | |
| Full fine-tuningBackbone=ViT-S/16, Trainable Parameters (%)=100%2026.03 | 79.6 | — | — | — | |
| DCBM-GDINOBackbone=CLIP ViT-B/162025.10 | 79.6 | — | — | — | |
| DCBM-MASKRCNNBackbone=CLIP ViT-B/162025.10 | 79.6 | — | — | — | |
| HyperfluxBackbone=ResNet-50, Pruning ratio=90.0%2025.04 | 79.58 | — | — | — | |
| PODSSelection Ratio=70%, Backbone=ResNet-182026.05 | 79.4 | — | — | — | |
| DCBM-SAM2Backbone=CLIP ViT-B/162025.10 | 79.4 | — | — | — | |
| HyperfluxBackbone=ResNet-50, Pruning ratio=95.0%2025.04 | 79.23 | — | — | — | |
| PODSSelection Ratio=50%, Backbone=ResNet-182026.05 | 79.1 | — | — | — | |
| GraNet (si = 0)Backbone=ResNet-50, Pruning ratio=90.0%2025.04 | 79.09 | — | — | — | |
| UGIESPruning Ratio=30%, Backbone=ResNet-182026.06 | 78.9 | — | — | — | |
| VISRegBackbone=ViT-B/16, Pre-training Epochs=400, Training Heuristics=w/o heuristics, Linear Probe Training Epochs=102026.06 | 78.8 | — | — | — | |
| GraNet (si = 0)Backbone=ResNet-50, Pruning ratio=95.0%2025.04 | 78.71 | — | — | — | |
| Dyn. Sel. Aug.Selection Ratio=70%, Backbone=ResNet-182026.05 | 78.6 | — | — | — | |
| UGIESPruning Ratio=50%, Backbone=ResNet-182026.06 | 78.6 | — | — | — | |
| DivBS†Pruning Ratio=30%, Backbone=ResNet-182026.06 | 78.5 | — | — | — | |
| S-TransformerT=42026.03 | 78.4 | — | — | — | |
| PODSSelection Ratio=30%, Backbone=ResNet-182026.05 | 78.4 | — | — | — | |
| GMPBackbone=ResNet-50, Pruning ratio=90.0%2025.04 | 78.39 | — | — | — | |
| GMPBackbone=ResNet-50, Pruning ratio=95.0%2025.04 | 78.38 | — | — | — | |
| ResNet-50 (Dense)Backbone=ResNet-50, Pruning ratio=0.0%2025.04 | 78.32 | — | — | — | |
| RS2Pruning Ratio=30%, Backbone=ResNet-182026.06 | 78.3 | — | — | — | |
| SpikformerT=42026.03 | 78.21 | — | — | — | |
| SpikingformerT=42026.03 | 78.21 | — | — | — | |
| Full DatasetSelection Ratio=100%, Backbone=ResNet-182026.05 | 78.2 | — | — | — | |
| InfoBatchSelection Ratio=70%, Backbone=ResNet-182026.05 | 78.2 | — | — | — | |
| InfoBatch*Pruning Ratio=30%, Backbone=ResNet-182026.06 | 78.2 | — | — | — | |
| DivBS†Pruning Ratio=50%, Backbone=ResNet-182026.06 | 78.2 | — | — | — | |
| Full DataPruning Ratio=0%, Backbone=ResNet-182026.06 | 78.2 | — | — | — | |
| Full datasetBackbone=ResNet-182026.03 | 78.19 | — | — | — | |
| InfoBatchSelection Ratio=50%, Backbone=ResNet-182026.05 | 78.1 | — | — | — | |
| InfoBatch*Pruning Ratio=50%, Backbone=ResNet-182026.06 | 78.1 | — | — | — | |
| RigLBackbone=ResNet-50, Pruning ratio=90.0%2025.04 | 78.04 | — | — | — | |
| GraNet (si = 0)Backbone=ResNet-50, Pruning ratio=98.0%2025.04 | 78.01 | — | — | — | |
| MAEBackbone=ViT-L/16, Pre-training Epochs=1600, Training Heuristics=w/o heuristics, Linear Probe Training Epochs=102026.06 | 78 | — | — | — | |
| CCT-7/3×2Architecture Type=Compact Convolutional ViT2026.03 | 77.72 | — | — | — | |
| HyperfluxBackbone=ResNet-50, Pruning ratio=98.0%2025.04 | 77.7 | — | — | — | |
| RL-SelectorSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.6 | — | — | — | |
| Dyn. Sel. Aug.Selection Ratio=50%, Backbone=ResNet-182026.05 | 77.6 | — | — | — | |
| RS2Pruning Ratio=50%, Backbone=ResNet-182026.06 | 77.6 | — | — | — | |
| UGIESPruning Ratio=70%, Backbone=ResNet-182026.06 | 77.6 | — | — | — | |
| MoSoSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.5 | — | — | — | |
| DUALSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.4 | — | — | — | |
| Label-Free-CBMBackbone=CLIP ViT-B/162025.10 | 77.4 | — | — | — | |
| RigLBackbone=ResNet-50, Pruning ratio=95.0%2025.04 | 77.39 | — | — | — | |
| Moderate-DSSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.3 | — | — | — | |
| Random*Selection Ratio=70%, Backbone=ResNet-182026.05 | 77.3 | — | — | — | |
| UCBSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.3 | — | — | — | |
| UCBPruning Ratio=30%, Backbone=ResNet-182026.06 | 77.3 | — | — | — | |
| EL2NSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.2 | — | — | — | |
| DPSelection Ratio=70%, Backbone=ResNet-182026.05 | 77.2 | — | — | — | |
| EL2NPruning Ratio=30%, Backbone=ResNet-182026.06 | 77.2 | — | — | — | |
| Dataset PruningPruning Ratio=30%, Backbone=ResNet-182026.06 | 77.2 | — | — | — | |
| DivBS†Pruning Ratio=70%, Backbone=ResNet-182026.06 | 77.2 | — | — | — | |
| GMPBackbone=ResNet-50, Pruning ratio=98.0%2025.04 | 77.16 | — | — | — | |
| FP32Model=SigLIP2-B/16, Evaluation Protocol=Zero-shot2025.10 | 77.1 | — | — | — | |
| CCSPruning Ratio=30%, Backbone=ResNet-182026.06 | 77.1 | — | — | — | |
| HAViTArchitecture Type=Modified ViT2026.03 | 77.07 | — | — | — | |
| CLIP-SelSelection Ratio=70%, Backbone=ResNet-182026.05 | 77 | — | — | — | |
| AUMPruning Ratio=30%, Backbone=ResNet-182026.06 | 76.9 | — | — | — | |
| FP32Model=OpenCLIP-B/16, Evaluation Protocol=Zero-shot2025.10 | 76.82 | — | — | — | |
| Self-sup. proto.Selection Ratio=70%, Backbone=ResNet-182026.05 | 76.8 | — | — | — | |
| MoSo†Pruning Ratio=30%, Backbone=ResNet-182026.06 | 76.7 | — | — | — | |
| ResNet-110Architecture Type=CNN2026.03 | 76.63 | — | — | — |