Table 2 Performance of I estimators for t-student and Gaussian samples (Relative Error in %)
From: Spatio-chromatic information available from different neural layers via Gaussianization
Shape | # Samples | Eq. (7) | Eq. (6) | KL+offset [81] | KL [79] | k-NN [80] |
|---|---|---|---|---|---|---|
T-STUDENT source/Ideal Infomax System | ||||||
ν = 3 | 1⋅103 | 24 ± 7 | 40 ± 30 | 58 ± 2 | 278 ± 2 | 239 ± 1 |
4⋅103 | 5 ± 2 | 16 ± 7 | 56 ± 1 | 259 ± 9 | 225 ± 1 | |
1⋅104 | 4 ± 2 | 16 ± 6 | 55 ± 1 | 246 ± 1 | 215 ± 1 | |
4⋅104 | 12 ± 2 | 18 ± 4 | 53 ± 1 | 229 ± 1 | 203 ± 1 | |
ν = 4 | 1⋅103 | 17 ± 8 | 25 ± 15 | 43 ± 2 | 255 ± 1 | 222 ± 1 |
4⋅103 | 3 ± 3 | 16 ± 8 | 43 ± 1 | 238 ± 1 | 210 ± 1 | |
1⋅104 | 7 ± 3 | 15 ± 6 | 41.1 ± 0.5 | 224 ± 1 | 199 ± 1 | |
4⋅104 | 17 ± 2 | 21 ± 7 | 39.4 ± 0.2 | 208 ± 1 | 188 ± 1 | |
ν = 100 | 1⋅103 | 15 ± 7 | 25 ± 14 | 20 ± 1 | 203 ± 1 | 185 ± 1 |
4⋅103 | 5 ± 2 | 18 ± 4 | 21.4 ± 0.7 | 187 ± 1 | 173 ± 1 | |
1⋅104 | 9 ± 2 | 21 ± 4 | 21.1 ± 0.1 | 174 ± 1 | 164 ± 1 | |
4⋅104 | 23 ± 3 | 29 ± 4 | 20.4 ± 0.1 | 160 ± 1 | 153 ± 1 | |
GAUSSIAN source/Trivial System | ||||||
1⋅103 | 24 ± 1 | 12 ± 2 | 6.0 ± 0.2 | 91.5 ± 0.2 | 95.6 ± 0.1 | |
4⋅103 | 16 ± 3 | 9 ± 3 | 6.4 ± 0.4 | 90.4 ± 0.4 | 95.1 ± 0.2 | |
1⋅104 | 10 ± 2 | 4 ± 1 | 6.4 ± 0.2 | 88.5 ± 0.2 | 93.8 ± 0.2 | |
4⋅104 | 7 ± 3 | 3 ± 2 | 6.1 ± 0.2 | 86.3 ± 0.2 | 92.3 ± 0.2 | |