Graph Scattering Networks with Adaptive Diffusion Kernels
Keywords: graph neural networks, graph scattering transform, deep learning, stability
TL;DR: We introduce a mathematically sound framework for applying adaptive diffusion kernels to the graph scattering transform.
Abstract: Scattering networks are deep convolutional architectures that use predefined wavelets for feature extraction and representation. They have proven effective for classification tasks, especially when training data is scarce, where traditional deep learning methods struggle. In this work, we introduce and develop a mathematically sound framework for applying adaptive kernels to diffusion wavelets in graph scattering networks. Stability guarantees with respect to input perturbations are provided. A specific construction of adaptive kernels is presented and applied with continuous diffusion to perform graph classification tasks on benchmark datasets. Our model consistently outperforms traditional graph scattering networks with predefined wavelets, both in scenarios with limited and abundant training data.
Primary Area: learning on graphs and other geometries & topologies
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 14139
Loading