Skip to main content

Table 3 This table illustrates the accuracy of deep learning models under varying sizes of datasets and quantities of perturbed points

From: Attack based on data: a novel perspective to attack sensitive points directly

Dataset name

Accuracy

Proportion

Number of perturbed points

2

3

5

7

10

Adiac

83.12

1.0

59.56

61.95

34.34

42.34

32.25

0.8

59.9

62.05

35.27

43.18

32.48

0.5

68.57

66.79

35.79

43.94

35.25

0.3

68.8

67.26

45.78

43.99

39.9

Car

93.33

1.0

83.34

84.26

73.87

79.48

88.5

0.8

92.28

85.07

82.19

80.27

89.11

0.5

92.73

92.75

85.59

80.49

89.92

0.3

93.33

93.33

90

90

90

ECGFiveDays

96.17

1.0

89.61

79.37

82.37

73.46

62.12

0.8

90.19

83.73

85.68

78.8

69.04

0.5

90.48

89.66

85.94

79.17

73.12

0.3

92.92

89.9

86.06

83.39

81.65

FaceAll

85.50

1.0

81.74

77.09

74.95

78.73

81.6

0.8

82.4

80.16

82.46

78.75

81.65

0.5

83.21

81.06

82.68

78.89

81.89

0.3

84.2

83.79

83.31

83.02

82.43

FISH

97.71

1.0

97.64

79.51

93.9

84.57

89.05

0.8

97.88

80.22

94.43

90.24

89.82

0.5

98.02

88.97

96.64

90.63

89.93

0.3

98.29

98.29

97.14

93.71

90.86

Meat

98.33

1.0

77.73

79.95

73.74

67.07

58.38

0.8

78.42

80.83

74.56

67.87

63.68

0.5

82.95

81.24

74.63

74.87

65.98

0.3

91.67

81.67

75

75

75

MedicalImages

76.18

1.0

64.29

66.22

59.95

54.67

50.25

0.8

65.23

70.87

60.48

55.38

50.99

0.5

75.05

71.82

60.67

56.14

54.85

0.3

75.39

72.5

65.26

61.05

56.97

PhalangesOutlinesCorrect

85.66

1.0

77.41

68.75

49.22

46.01

46.97

0.8

78.35

69.01

58.8

46.05

47.89

0.5

80.92

69.76

59.56

51.5

48.42

0.3

81.35

72.96

60.02

51.98

48.95

Strawberry

96.25

1.0

86.07

81.26

77.76

77.09

78.84

0.8

93.52

91.15

81.25

78.03

79.49

0.5

94.22

91.8

85.91

82.77

80.22

0.3

94.78

91.84

88.09

83.2

80.91

SwedishLeaf

95.36

1.0

94.53

88.69

82.86

79.27

77.05

0.8

94.6

89.16

92.11

89.25

77.68

0.5

94.72

89.61

92.58

89.42

82.53

0.3

94.72

94.4

92.96

90.08

86.72

  1. “accuracy” refers the accuracy of the dataset before perturbed, “proportion” refers to the proportion of the dataset available for use during attacks, “number of perturbed points” refers the quantity of points that have been perturbed.(\(\beta\) is set to 0.1)