Multiview Variational Graph Autoencoders For Canonical Correlation Analysis

We present a novel Multiview Canonical Correlation Analysis model based on a variational approach. This is the first non linear model able to take into account some a priori graph- based geometric constraints while being scalable for process- ing large scale datasets with multiple views. It is based on an autoencoder architecture making use of Graph Convolu- tional Neural network models. We experiment our approach on classification, clustering and recommendation tasks. The algorithm is competitive among multiview models taking ac- count geometric information while remaining more scalable.