Table 4.

Breast cancer data: SVM is based on stacked raw data with two views

Method	Average Error (Std.Dev) (%)
SVM	39.02 (4.77)
Deep CCA + SVM	38.57 (5.40)
Sparse CCA + SVM	40.94 (4.24)
MOMA	44.51 (3.90)
MOMA + SVM	39.46 (5.67)
Random Forest	40.36 (5.28)
iDeepViewLearn with selected top 10% features	39.02 (5.03)
iDeepViewLearn with selected top 20% features	39.02 (5.03)
iDeepViewLearn with selected top 10% stacked features	39.11 (4.82)
iDeepViewLearn with selected top 20% stacked features	39.38 (5.55)

Deep CCA + SVM is a training SVM based on the last layer of Deep CCA. iDeepViewLearn with selected top $10\%$ features reconstructs the original views with only $10\%$ of the features and obtains a test classification error based on a shared low-dimensional representation trained on data with only $10\%$ of the features. Similar to iDeepViewLearn with selected top $20\%$. (The mean error of two views is reported for MOMA; MOMA + SVM means combining the feature selection part of MOMA and SVM)