Skip to main content

Table 4 Number of noisy data selections by the AL methods in each iteration

From: A predicted-loss based active learning approach for robust cancer pathology image analysis in the workplace

Iteration

LC

Entropy

BALD

LL

RS

Core-set

Proposed

1

0.0

0.0

0.0

0.0

0.0

0.0

0.0

2

38.2

191.0

114.6

878.4

60.3

326.3

15.2

3

132.0

660.0

396.0

626.4

57.8

55.0

28.2

4

86.1

430.6

258.3

667.8

61.8

17.7

20.4

5

114.3

571.4

342.8

558.4

57.5

30.3

36.0

6

74.6

373.0

223.8

161.2

53.3

101.3

25.8

7

50.7

253.4

152.0

115.2

60.0

9.3

31.0

8

148.2

740.8

444.5

35

65.5

7.7

33.8

9

88.5

442.6

265.6

62

60.5

7.7

37.0

10

67.0

335.2

201.1

73.2

59.0

6.0

36.0

Mean

88.8

444.2

266.5

353.1

59.5

62.36

29.3

Total

799.6

3998

2398.8

3177.6

535.5

561.3

263.4

N-Ratio

8.9%

44.4%

26.7%

35.3%

6.0%

6.2%

2.93%

  1. LCĀ Least confidence, BALD Bayesian active learning by disagreement, LLĀ Learning loss, RSĀ Random sampling, N-RatioĀ Noise-ratio