On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features
While Graph Neural Networks (GNNs) have recently become the 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...
| Prif Awduron: | , , , , , |
|---|---|
| Fformat: | Conference item |
| Iaith: | English |
| Cyhoeddwyd: |
Proceedings of Machine Learning Research
2022
|
| _version_ | 1826311477131214848 |
|---|---|
| author | Rossi, E Kenlay, H Gorinova, MI Chamberlain, BP Dong, X Bronstein, M |
| author_facet | Rossi, E Kenlay, H Gorinova, MI Chamberlain, BP Dong, X Bronstein, M |
| author_sort | Rossi, E |
| collection | OXFORD |
| description | While Graph Neural Networks (GNNs) have recently become the 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 minimization of 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 seven common node-classification benchmarks 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 ~2.5M nodes and ~23M edges on a single GPU. The code is available at https://github.com/twitter-research/feature-propagation. |
| first_indexed | 2024-03-07T08:10:24Z |
| format | Conference item |
| id | oxford-uuid:f35d56fb-0e47-48ce-9f55-f7b97593897d |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:10:24Z |
| publishDate | 2022 |
| publisher | Proceedings of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:f35d56fb-0e47-48ce-9f55-f7b97593897d2023-11-16T16:18:27ZOn the unreasonable effectiveness of feature propagation in learning on graphs with missing node featuresConference itemhttp://purl.org/coar/resource_type/c_5794uuid:f35d56fb-0e47-48ce-9f55-f7b97593897dEnglishSymplectic ElementsProceedings of Machine Learning Research2022Rossi, EKenlay, HGorinova, MIChamberlain, BPDong, XBronstein, MWhile Graph Neural Networks (GNNs) have recently become the 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 minimization of 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 seven common node-classification benchmarks 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 ~2.5M nodes and ~23M edges on a single GPU. The code is available at https://github.com/twitter-research/feature-propagation. |
| spellingShingle | Rossi, E Kenlay, H Gorinova, MI Chamberlain, BP Dong, X Bronstein, M On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| title | On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| title_full | On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| title_fullStr | On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| title_full_unstemmed | On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| title_short | On the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| title_sort | on the unreasonable effectiveness of feature propagation in learning on graphs with missing node features |
| work_keys_str_mv | AT rossie ontheunreasonableeffectivenessoffeaturepropagationinlearningongraphswithmissingnodefeatures AT kenlayh ontheunreasonableeffectivenessoffeaturepropagationinlearningongraphswithmissingnodefeatures AT gorinovami ontheunreasonableeffectivenessoffeaturepropagationinlearningongraphswithmissingnodefeatures AT chamberlainbp ontheunreasonableeffectivenessoffeaturepropagationinlearningongraphswithmissingnodefeatures AT dongx ontheunreasonableeffectivenessoffeaturepropagationinlearningongraphswithmissingnodefeatures AT bronsteinm ontheunreasonableeffectivenessoffeaturepropagationinlearningongraphswithmissingnodefeatures |