Skip to main content
EURASIP Journal on Advances in Signal Processing

Table 2 Standard deviation estimates (in nats) with respect to data processing

From: An experimental study of neural estimators of the mutual information between random vectors modeling power spectrum features

Estimators

true MI

Raw

Logarithm

Squaring

Scaling

Shift

NWJ

4

0.34

0.23

0.24

0.21

0.27

12

4.93

21.60

7.20

3.53

3.21

20

17.86

8.06

5.26

29.52

29.68

NCE

4

0.16

0.15

0.16

0.16

0.15

12

0.03

0.03

0.04

0.04

0.03

20

0.01

0.01

0.01

0.01

0.01

MINE

4

0.25

0.24

0.22

0.20

0.31

12

1.08

1.02

0.96

1.07

1.34

20

2.44

2.30

2.45

2.72

2.32

SMILE

4

0.29

0.23

0.27

0.26

0.28

12

0.71

0.49

0.63

0.71

0.64

20

1.34

1.04

1.34

1.13

1.47

FLOW

4

0.33

0.24

0.79

0.43

0.32

12

0.38

0.32

0.85

0.44

0.37

20

0.38

0.36

0.85

0.44

0.40

Combined

4

0.29

0.23

0.27

0.26

0.28

12

0.45

0.37

0.84

0.44

0.64

20

0.38

0.36

0.85

0.44

0.40

  1. Best results are indicated in bold, except for the standard deviation of the NCE estimator (indicated in italics), which restricts the deviation at the cost of upper bounding the estimator
Back to article page