Learnable graph convolutional network and feature fusion for multi-view learning

In practical applications, multi-view data depicting objects from assorted perspectives can facilitate the accuracy increase of learning algorithms. However, given data, there is limited work for discriminative node relationships and graph information simultaneously via convolutional network that has drawn attention considerable researchers in recent years. Most existing methods only consider weighted sum adjacency matrices, yet a joint neural both feature fusion still under-explored. To cope with these issues, this paper proposes deep framework called Learnable Graph Convolutional Network Feature Fusion (LGCN-FF), consisting two modules: learnable network. The former aims to learn an underlying representation heterogeneous views, while latter explores more weights parametric activation function dubbed Differentiable Shrinkage Activation (DSA) function. proposed LGCN-FF validated be superior various state-of-the-art semi-supervised classification.