Table 4:

Noisy MNIST data: SVM was implemented on the stacked data. For Deep CCA + SVM, we trained SVM on the combined outputs (from view 1 and view 2) of the last layer of Deep CCA. For Deep IDA + NCC, we implemented the Nearest Centroid Classification on the combined outputs (from view 1 and view 2) of the last layer of Deep IDA. For Deep IDA + SVM, we trained SVM on the combined outputs (from view 1 and view 2) of the last layer of Deep IDA.

Method	Accuracy (%)
SVM (combined view 1 and 2)	93.75
Deep CCA + SVM	97.01
Deep IDA + NCC	97.74
Deep IDA + SVM	99.15
RKCCA + SVM	91.79