Deconvolutional Networks on Graph Data
This addresses a specific inverse problem in graph learning for researchers and practitioners, but appears incremental as it builds on existing GCN methods.
The paper tackles the inverse problem of reconstructing input graph signals from smoothed graph representations produced by Graph Convolutional Networks, proposing a Graph Deconvolutional Network that achieves effectiveness in tasks like graph feature imputation and structure generation.
In this paper, we consider an inverse problem in graph learning domain -- ``given the graph representations smoothed by Graph Convolutional Network (GCN), how can we reconstruct the input graph signal?" We propose Graph Deconvolutional Network (GDN) and motivate the design of GDN via a combination of inverse filters in spectral domain and de-noising layers in wavelet domain, as the inverse operation results in a high frequency amplifier and may amplify the noise. We demonstrate the effectiveness of the proposed method on several tasks including graph feature imputation and graph structure generation.