Table 3 As in Table 3, comparing Robust Sparse CCA to other alternatives in the “Sparse High-dimensional 2 design”
Method | No contamination | t-distribution | Contamination | ||||||
|---|---|---|---|---|---|---|---|---|---|
\(\bar {\theta }(\hat {\mathbf {A}},\mathbf {A})\) | TPR | TNR | \(\bar {\theta }(\hat {\mathbf {A}},\mathbf {A})\) | TPR | TNR | \(\bar {\theta }(\hat {\mathbf {A}},\mathbf {A})\) | TPR | TNR | |
Sparse CCA of [15] | 0.93 | 1.00 | 0.93 | 1.41 | 0.94 | 0.72 | 1.28 | 0.89 | 0.00 |
Sparse CCA of [16] | 0.79 | 0.65 | 1.00 | 1.16 | 0.30 | 0.92 | 1.57 | 0.00 | 0.00 |
Sparse CCA of [17] | 0.44 | 1.00 | 0.08 | 1.01 | 1.00 | 0.02 | 1.25 | 1.00 | 0.00 |
Sparse CCA on pre-processed data | 0.58 | 0.92 | 0.79 | 0.72 | 0.88 | 0.74 | 1.36 | 0.74 | 0.25 |
Sparse CCA with robust initialization | 1.07 | 0.32 | 1.00 | 1.13 | 0.24 | 1.00 | 1.25 | 0.38 | 0.97 |
Robust Sparse CCA | 0.59 | 0.87 | 0.87 | 0.60 | 0.94 | 0.89 | 0.84 | 0.97 | 0.82 |