Table 3 Summary of effectiveness of each active learning strategy in across different key characteristics
From: Deep active learning for classifying cancer pathology reports
| Â | Overall micro (Large) | Overall macro (Large) | Overall micro (Small) | Overall macro (Small) | Single iteration micro (Small) | Single iteration macro (Small) | Compute cost |
|---|---|---|---|---|---|---|---|
Least Con. | High | High | Med | High | Low | Med | Low |
Entropy Sam. | High | High | Med | High | Low | Med | Low |
Ratio Sam. | High | High | High | Med | High | High | Low |
Marginal Sam. | High | High | High | Med | High | High | Low |
Euclidian C. (D.S.) | V. Low | V. Low | V. Low | V. Low | V. Low | V. Low | Med |
Cosine C. (D.S.) | V. Low | V. Low | V. Low | V. Low | V. Low | V. Low | Med |
Information Den. | High | High | Low | Low | Low | Low | Med |
Meta Learning | Med | Med | Low | Low | Low | Low | Med |
QBC—S.S. | High | High | Med | High | Low | High | High |
QBC—VE | High | High | Med | High | Low | High | High |
QBC—KL-D | High | High | Med | High | Low | High | High |