Active Learning Image Classification on CIFAR-100
81.86AccuracyEnsemble AL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Ensemble ALBackbone=ResNet-18, Candidate Model Used=Yes2026.05 | 81.86 | 0 | — | |
| Low Confidence (LC)Backbone=MobileNetV2, Candidate Model Used=No2026.05 | 73.79 | 0.25 | 6.63 | |
| High Confidence and Least Confidence (HCLC)Backbone=MobileNetV2, Candidate Model Used=No2026.05 | 73.79 | 0.25 | 6.63 | |
| High Confidence and Least Confidence (HCLC)Backbone=ResNet-18, Candidate Model Used=No2026.05 | 73.42 | 0.1 | 3.88 | |
| Low Confidence (LC)Backbone=ResNet-18, Candidate Model Used=No2026.05 | 73.24 | 0.1 | 3.88 | |
| High Confidence (HC)Backbone=MobileNetV2, Candidate Model Used=No2026.05 | 72.82 | 0.25 | 6.63 | |
| High Confidence (HC)Backbone=ResNet-18, Candidate Model Used=No2026.05 | 71.97 | 0.1 | 3.88 | |
| BADGEBackbone=DenseNet-121, Candidate Model Used=Yes2026.05 | 71.82 | 0 | 7.9 | |
| Low Confidence (LC)Backbone=DenseNet-121, Candidate Model Used=No2026.05 | 71.65 | 0.19 | 5.27 | |
| Passive Batch Injection TrainingBackbone=ResNet-56, Candidate Model Used=No2026.05 | 71.6 | — | — | |
| High Confidence and Least Confidence (HCLC)Backbone=DenseNet-121, Candidate Model Used=No2026.05 | 71.33 | 0.19 | 5.27 | |
| PrACBackbone=ResNet-56, Candidate Model Used=Yes2026.05 | 71.2 | 0 | — | |
| High Confidence (HC)Backbone=DenseNet-121, Candidate Model Used=No2026.05 | 70.98 | 0.19 | 5.27 | |
| Linear Feature DisentanglementBackbone=ResNet-56, Candidate Model Used=No2026.05 | 70.73 | — | — | |
| active-iNASBackbone=ResNet-18, Candidate Model Used=Yes2026.05 | 70.5 | 0 | — | |
| Coreset – WSMBackbone=VGG-16, Candidate Model Used=Yes2026.05 | 70 | 0 | — | |
| GlisterBackbone=DenseNet-121, Candidate Model Used=Yes2026.05 | 68.36 | 0 | 4.96 | |
| PruneFuseBackbone=ResNet-56, Candidate Model Used=Yes2026.05 | 67.87 | 0 | — | |
| FF-Active (variant)Backbone=VGG-16, Candidate Model Used=Yes2026.05 | 67 | 0 | — | |
| Low Confidence (LC)Backbone=VGG-16, Candidate Model Used=No2026.05 | 66.46 | 0.57 | 9.1 | |
| High Confidence and Least Confidence (HCLC)Backbone=ResNet-56, Candidate Model Used=No2026.05 | 66.39 | 0.31 | 8.51 | |
| High Confidence (HC)Backbone=ResNet-56, Candidate Model Used=No2026.05 | 66.3 | 0.31 | 8.51 | |
| CoresetBackbone=MobileNetV2, Candidate Model Used=Yes2026.05 | 66.27 | 0 | — | |
| Low Confidence (LC)Backbone=ResNet-56, Candidate Model Used=No2026.05 | 66.22 | 0.31 | 8.51 | |
| High Confidence and Least Confidence (HCLC)Backbone=VGG-16, Candidate Model Used=No2026.05 | 66.11 | 0.57 | 9.1 | |
| FF-ActiveBackbone=VGG-16, Candidate Model Used=Yes2026.05 | 66 | 0 | — | |
| BAITBackbone=DenseNet-121, Candidate Model Used=Yes2026.05 | 65.91 | 0 | 9.75 | |
| Coreset – FSMBackbone=VGG-16, Candidate Model Used=Yes2026.05 | 65 | 0 | — | |
| High Confidence (HC)Backbone=VGG-16, Candidate Model Used=No2026.05 | 64.87 | 0.57 | 9.1 | |
| GlisterBackbone=MobileNetV2, Candidate Model Used=Yes2026.05 | 64.78 | 0 | — | |
| LAL-IGradV-VAEBackbone=MobileNetV2, Candidate Model Used=Yes2026.05 | 64.5 | 0 | — | |
| Moderate CoresetBackbone=MobileNetV2, Candidate Model Used=Yes2026.05 | 64.45 | 0 | — | |
| Influence SelectionBackbone=ResNet-18, Candidate Model Used=Yes2026.05 | 64 | 0 | — | |
| TLCBackbone=DenseNet-121, Candidate Model Used=Yes2026.05 | 62.05 | 0 | — | |
| SuperconvergenceBackbone=ResNet-56, Candidate Model Used=No2026.05 | 60.8 | — | — | |
| Bayesian Neural NetworksBackbone=DenseNet-121, Candidate Model Used=Yes2026.05 | 60 | 0 | — | |
| Yun et al.Backbone=ResNet-18, Candidate Model Used=Yes, Budget=10K2026.05 | 59.73 | 0 | — | |
| Low Confidence (LC)Backbone=DenseNet-121, Candidate Model Used=No, Budget=10K2026.05 | 59.03 | 0.19 | 0.4 | |
| Low Confidence (LC)Backbone=ResNet-18, Candidate Model Used=No, Budget=10K2026.05 | 59.01 | 0.1 | 3.88 | |
| Rakesh & JainBackbone=DenseNet-121, Candidate Model Used=Yes, Budget=10K2026.05 | 59 | — | — | |
| Adversarial SamplingBackbone=ResNet-56, Candidate Model Used=Yes2026.05 | 57.5 | 0 | — | |
| ST-CoNALBackbone=ResNet-18, Candidate Model Used=Yes, Budget=10K2026.05 | 57.49 | 0 | — | |
| Yun et al.Backbone=DenseNet-121, Candidate Model Used=Yes, Budget=10K2026.05 | 57.43 | 0 | — | |
| CoreGCNBackbone=ResNet-18, Candidate Model Used=Yes, Budget=10K2026.05 | 56.5 | 0 | — | |
| UncertainGCNBackbone=ResNet-18, Candidate Model Used=Yes, Budget=10K2026.05 | 55 | 0 | — | |
| VAALBackbone=VGG-16, Candidate Model Used=Yes2026.05 | 47.6 | 0 | — |