2019/08/07 by Jiwoong Park, Park, Jiwoong, Minsik Lee +7 · 4 citations
Computer Science · Biochemistry, Genetics and Molecular Biology · Physics and Astronomy · #Advanced Graph Neural Networks #Bioinformatics and Genomic Networks #Complex Network Analysis Techniques
paper · pdf · doi:10.48550/arxiv.1908.02441
We propose a symmetric graph convolutional autoencoder which produces a\nlow-dimensional latent representation from a graph. In contrast to the existing\ngraph autoencoders with asymmetric decoder parts, the proposed autoencoder has\na newly designed decoder which builds a completely symmetric autoencoder form.\nFor the reconstruction of node features, the decoder is designed based on\nLaplacian sharpening as the counterpart of Laplacian smoothing of the encoder,\nwhich allows utilizing the graph structure in the whole processes of the\nproposed autoencoder architecture. In order to prevent the numerical\ninstability of the network caused by the Laplacian sharpening introduction, we\nfurther propose a new numerically stable form of the Laplacian sharpening by\nincorporating the signed graphs. In addition, a new cost function which finds a\nlatent representation and a latent affinity matrix simultaneously is devised to\nboost the performance of image clustering tasks. The experimental results on\nclustering, link prediction and visualization tasks strongly support that the\nproposed model is stable and outperforms various state-of-the-art algorithms.\n