Table 1 Accuracy of active learning methods on CIFAR10.
From: Unlabeled data selection for active learning in image classification
Method | Cycle 2 (%) | Cycle 4 (%) | Cycle 6 (%) | Cycle 8 |
|---|---|---|---|---|
Random sampling | 82.86 | 86.54 | 88.29 | 89.86% |
BALD | 85.55 | 89.37 | 91.16 | 92.43 |
Similarity-based selection | 82.68 | 85.83 | 88.34 | 89.89 |
Prediction probability-based selection | 87.28 | 90.17 | 92.67 | 93.56 |
QBC (query by committee) | 83.75 | 87.37 | 89.15 | 90.62 |
EMC(expected model change) | 84.11 | 88.28 | 90.05 | 91.56 |
Bayesian generative AL | 85.23 | 89.02 | 91.17 | 92.75 |
