Table 1.
Estimation of the information of a subset of regions using linear kernels along with ν-MKL and lp-norm MKL for the simulated dataset. The metrics used to determine the amount of information of the regions by means of ν-MKL (mean of the normalized γ values) and lp-norm MKL (kernel weights’ mean) as well as their selection frequencies for each algorithm are reported. Both the normalized γ values and the kernel weights have been scaled so that their maximum values equal 1 to make the comparison easier. These coefficients are contrasted against the accuracy rates achieved by these regions using a linear SVM.
| Region | Linear SVM | ν-MKL | lp-norm MKL | ||
|---|---|---|---|---|---|
|
| |||||
| Acc. Rate | Sel. Freq. | Normalized γ | Sel. Freq. | Kernel Weights | |
| Square 26 | 0.81 | 1 | 1.00 | 1 | 0.91 |
| Square 46 | 0.78 | 1 | 0.95 | 1 | 0.91 |
| Square 32 | 0.77 | 1 | 0.99 | 1 | 1.00 |
| Square 77 | 0.76 | 1 | 0.91 | 1 | 0.72 |
| Square 29 | 0.76 | 1 | 0.76 | 1 | 0.67 |
| Square 23 | 0.76 | 1 | 0.71 | 1 | 0.81 |
| Square 12 | 0.75 | 1 | 0.75 | 1 | 0.53 |
| Square 57 | 0.69 | 1 | 0.54 | 0.50 | 0.58 |
| Square 51 | 0.68 | 1 | 0.52 | 1 | 0.34 |
| Square 30 | 0.67 | 1 | 0.24 | 0.50 | 0.34 |
| Square 107 | 0.63 | 0.60 | 0.08 | 0.60 | 0.30 |
| Square 13 | 0.60 | 0.60 | 0.09 | 0.50 | 0.38 |
| Square 44 | 0.57 | 0.30 | 0.13 | 0.90 | 0.29 |
| Square 37 | 0.56 | 0.10 | 0.09 | 0.90 | 0.24 |
| Square 20 | 0.54 | 0.10 | 0.07 | 0.80 | 0.22 |