Graph classification Gaussian processes via spectral features
Graph classification aims to categorise graphs based on their structure and node attributes. In this work, we propose to tackle this task using tools from graph signal processing by deriving spectral features, which we then use to design two variants of Gaussian process models for graph classificati...
| Main Authors: | , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Proceedings of Machine Learning Research
2023
|
| _version_ | 1826311384777883648 |
|---|---|
| author | Opolka, FL Zhi, Y-C Liò, P Dong, X |
| author_facet | Opolka, FL Zhi, Y-C Liò, P Dong, X |
| author_sort | Opolka, FL |
| collection | OXFORD |
| description | Graph classification aims to categorise graphs based on their structure and node attributes. In this work, we propose to tackle this task using tools from graph signal processing by deriving spectral features, which we then use to design two variants of Gaussian process models for graph classification. The first variant uses spectral features based on the distribution of energy of a node feature signal over the spectrum of the graph. We show that even such a simple approach, having no learned parameters, can yield competitive performance compared to strong neural network and graph kernel baselines. A second, more sophisticated variant is designed to capture multi-scale and localised patterns in the graph by learning spectral graph wavelet filters, obtaining improved performance on synthetic and real-world data sets. Finally, we show that both models produce well calibrated uncertainty estimates, enabling reliable decision making based on the model predictions. |
| first_indexed | 2024-03-07T08:09:03Z |
| format | Conference item |
| id | oxford-uuid:13fedd3f-c70d-40f8-bd69-d8549e83bb01 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:09:03Z |
| publishDate | 2023 |
| publisher | Proceedings of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:13fedd3f-c70d-40f8-bd69-d8549e83bb012023-11-16T16:37:56ZGraph classification Gaussian processes via spectral featuresConference itemhttp://purl.org/coar/resource_type/c_5794uuid:13fedd3f-c70d-40f8-bd69-d8549e83bb01EnglishSymplectic ElementsProceedings of Machine Learning Research2023Opolka, FLZhi, Y-CLiò, PDong, XGraph classification aims to categorise graphs based on their structure and node attributes. In this work, we propose to tackle this task using tools from graph signal processing by deriving spectral features, which we then use to design two variants of Gaussian process models for graph classification. The first variant uses spectral features based on the distribution of energy of a node feature signal over the spectrum of the graph. We show that even such a simple approach, having no learned parameters, can yield competitive performance compared to strong neural network and graph kernel baselines. A second, more sophisticated variant is designed to capture multi-scale and localised patterns in the graph by learning spectral graph wavelet filters, obtaining improved performance on synthetic and real-world data sets. Finally, we show that both models produce well calibrated uncertainty estimates, enabling reliable decision making based on the model predictions. |
| spellingShingle | Opolka, FL Zhi, Y-C Liò, P Dong, X Graph classification Gaussian processes via spectral features |
| title | Graph classification Gaussian processes via spectral features |
| title_full | Graph classification Gaussian processes via spectral features |
| title_fullStr | Graph classification Gaussian processes via spectral features |
| title_full_unstemmed | Graph classification Gaussian processes via spectral features |
| title_short | Graph classification Gaussian processes via spectral features |
| title_sort | graph classification gaussian processes via spectral features |
| work_keys_str_mv | AT opolkafl graphclassificationgaussianprocessesviaspectralfeatures AT zhiyc graphclassificationgaussianprocessesviaspectralfeatures AT liop graphclassificationgaussianprocessesviaspectralfeatures AT dongx graphclassificationgaussianprocessesviaspectralfeatures |