Handling Missing Data via Max-Entropy Regularized Graph Autoencoder

Graph neural networks (GNNs) are popular weapons for modeling relational data. Existing GNNs not specified attribute-incomplete graphs, making missing attribute imputation a burning issue. Until recently, many works notice that coupled with spectral concentration, which means the spectrum obtained by concentrates on local part in domain, e.g., low-frequency due to oversmoothing As consequence, may be seriously flawed reconstructing graph attributes as concentration tends cause low precision. In this work, we present regularized autoencoder imputation, named MEGAE, aims at mitigating problem maximizing entropy. Notably, first method estimating entropy without eigen-decomposition of Laplacian matrix and provide theoretical upper error bound. A maximum regularization then acts latent space, directly increases Extensive experiments show MEGAE outperforms all other state-of-the-art methods variety benchmark datasets.