On the Unreasonable Effectiveness of Feature propagation in Learning on Graphs with Missing Node Features

Graph machine learning has recently become the de facto standard for modeling relational data.

However, many real-world applications require a strong assumption on the availability of the node or edge features of the graph.

We present a general approach for handling missing features in graph machine learning applications that is based on minimization of the dirichlet energy andleads to a diffusion-type differential equation on the graph.

The discretization of this equation produces a simple, fast and scalable algorithm which we call feature propagation.

We experimentally show that the proposed approach outperforms previous methods on seven common node-classificationbenchmarks and can withstand surprisingly high rates of missing features : on average we observe only around 4% relative accuracy drop when 99% of the features are missing.

Moreover, it takes only 10 seconds to run on a graph with nodes and edges on a single gpu.

Authors

Emanuele Rossi, Henry Kenlay, Maria I. Gorinova, Benjamin Paul Chamberlain, Xiaowen Dong, Michael Bronstein