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

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

Keywords: graph neural networks, missing features, graphs, feature propagation

Abstract:

While Graph Neural Networks (GNNs) have recently become the {\em de facto} standard for modeling relational data, they impose a strong assumption on the availability of the node or edge features of the graph. In many real-world applications, however,
features are only partially available; for example, in social networks, age and gender are available only for a small subset of users. We present a general approach for handling missing features in graph machine learning applications that is based on minimizing the Dirichlet energy and leads 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 six common node-classification benchmarks and can withstand surprisingly high rates of missing features: on average we observe only around 5% relative accuracy drop when 99% of the features are missing. Moreover, it takes only 10 seconds to run on a graph with ~2.5M nodes and ~123M edges on a single GPU.

One-sentence Summary: We propose a diffusion-based framework to learn on graphs with missing features, and we show that it can gracefully withstand surprisingly high rates of missing features