Search-Convolutional Neural Networks

We present a new deterministic relational model derived from convolutional neural networks. Search-Convolutional Neural Networks (SCNNs) extend the notion of convolution to graph search to construct a rich latent representation that extracts local behavior from general graph-structured data. Unlike other neural network models that take graph-structured data as input, SCNNs have a parameterization that is independent of input size, a property that enables transfer learning between datasets. SCNNs can be applied to a wide variety of prediction tasks, including node classification, community detection, and link prediction. Our results indicate that SCNNs can offer considerable lift over off-the-shelf classifiers and simple multilayer perceptrons, and comparable performance to state-of-the-art probabilistic graphical models at considerably lower computational cost.