BLIS-Net: Classifying and Analyzing Signals on Graphs

Charles Xu, Laney Goldman, Valentina Guo, Benjamin Hollander-Bodie, Maedee Trank-Greene, Ian Adelstein, Edward De Brouwer, Rex Ying, Smita Krishnaswamy, Michael Perlmutter
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:4537-4545, 2024.

Abstract

Graph neural networks (GNNs) have emerged as a powerful tool for tasks such as node classification and graph classification. However, much less work has been done on signal classification, where the data consists of many functions (referred to as signals) defined on the vertices of a single graph. These tasks require networks designed differently from those designed for traditional GNN tasks. Indeed, traditional GNNs rely on localized low-pass filters, and signals of interest may have intricate multi-frequency behavior and exhibit long range interactions. This motivates us to introduce the BLIS-Net (Bi-Lipschitz Scattering Net), a novel GNN that builds on the previously introduced geometric scattering transform. Our network is able to capture both local and global signal structure and is able to capture both low-frequency and high-frequency information. We make several crucial changes to the original geometric scattering architecture which we prove increase the ability of our network to capture information about the input signal and show that BLIS-Net achieves superior performance on both synthetic and real-world data sets based on traffic flow and fMRI data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-xu24c, title = {{BLIS}-{N}et: Classifying and Analyzing Signals on Graphs}, author = {Xu, Charles and Goldman, Laney and Guo, Valentina and Hollander-Bodie, Benjamin and Trank-Greene, Maedee and Adelstein, Ian and De Brouwer, Edward and Ying, Rex and Krishnaswamy, Smita and Perlmutter, Michael}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {4537--4545}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/xu24c/xu24c.pdf}, url = {https://proceedings.mlr.press/v238/xu24c.html}, abstract = {Graph neural networks (GNNs) have emerged as a powerful tool for tasks such as node classification and graph classification. However, much less work has been done on signal classification, where the data consists of many functions (referred to as signals) defined on the vertices of a single graph. These tasks require networks designed differently from those designed for traditional GNN tasks. Indeed, traditional GNNs rely on localized low-pass filters, and signals of interest may have intricate multi-frequency behavior and exhibit long range interactions. This motivates us to introduce the BLIS-Net (Bi-Lipschitz Scattering Net), a novel GNN that builds on the previously introduced geometric scattering transform. Our network is able to capture both local and global signal structure and is able to capture both low-frequency and high-frequency information. We make several crucial changes to the original geometric scattering architecture which we prove increase the ability of our network to capture information about the input signal and show that BLIS-Net achieves superior performance on both synthetic and real-world data sets based on traffic flow and fMRI data.} }
Endnote
%0 Conference Paper %T BLIS-Net: Classifying and Analyzing Signals on Graphs %A Charles Xu %A Laney Goldman %A Valentina Guo %A Benjamin Hollander-Bodie %A Maedee Trank-Greene %A Ian Adelstein %A Edward De Brouwer %A Rex Ying %A Smita Krishnaswamy %A Michael Perlmutter %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-xu24c %I PMLR %P 4537--4545 %U https://proceedings.mlr.press/v238/xu24c.html %V 238 %X Graph neural networks (GNNs) have emerged as a powerful tool for tasks such as node classification and graph classification. However, much less work has been done on signal classification, where the data consists of many functions (referred to as signals) defined on the vertices of a single graph. These tasks require networks designed differently from those designed for traditional GNN tasks. Indeed, traditional GNNs rely on localized low-pass filters, and signals of interest may have intricate multi-frequency behavior and exhibit long range interactions. This motivates us to introduce the BLIS-Net (Bi-Lipschitz Scattering Net), a novel GNN that builds on the previously introduced geometric scattering transform. Our network is able to capture both local and global signal structure and is able to capture both low-frequency and high-frequency information. We make several crucial changes to the original geometric scattering architecture which we prove increase the ability of our network to capture information about the input signal and show that BLIS-Net achieves superior performance on both synthetic and real-world data sets based on traffic flow and fMRI data.
APA
Xu, C., Goldman, L., Guo, V., Hollander-Bodie, B., Trank-Greene, M., Adelstein, I., De Brouwer, E., Ying, R., Krishnaswamy, S. & Perlmutter, M.. (2024). BLIS-Net: Classifying and Analyzing Signals on Graphs. Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:4537-4545 Available from https://proceedings.mlr.press/v238/xu24c.html.

Related Material