TY - JOUR
T1 - Learnable Filters for Geometric Scattering Modules
AU - Tong, Alexander
AU - Wenkel, Frederik
AU - Bhaskar, Dhananjay
AU - Macdonald, Kincaid
AU - Grady, Jackson
AU - Perlmutter, Michael
AU - Krishnaswamy, Smita
AU - Wolf, Guy
N1 - Publisher Copyright:
© 1991-2012 IEEE.
PY - 2024
Y1 - 2024
N2 - We propose a new graph neural network (GNN) module, based on relaxations of recently proposed geometric scattering transforms, which consist of a cascade of graph wavelet filters. Our learnable geometric scattering (LEGS) module enables adaptive tuning of the wavelets to encourage band-pass features to emerge in learned representations. The incorporation of our LEGS-module in GNNs enables the learning of longer-range graph relations compared to many popular GNNs, which often rely on encoding graph structure via smoothness or similarity between neighbors. Further, its wavelet priors result in simplified architectures with significantly fewer learned parameters compared to competing GNNs. We demonstrate the predictive performance of LEGS-based networks on graph classification benchmarks, as well as the descriptive quality of their learned features in biochemical graph data exploration tasks. Our results show that LEGS-based networks match or outperforms popular GNNs, as well as the original geometric scattering construction, on many datasets, in particular in biochemical domains, while retaining certain mathematical properties of handcrafted (non-learned) geometric scattering.
AB - We propose a new graph neural network (GNN) module, based on relaxations of recently proposed geometric scattering transforms, which consist of a cascade of graph wavelet filters. Our learnable geometric scattering (LEGS) module enables adaptive tuning of the wavelets to encourage band-pass features to emerge in learned representations. The incorporation of our LEGS-module in GNNs enables the learning of longer-range graph relations compared to many popular GNNs, which often rely on encoding graph structure via smoothness or similarity between neighbors. Further, its wavelet priors result in simplified architectures with significantly fewer learned parameters compared to competing GNNs. We demonstrate the predictive performance of LEGS-based networks on graph classification benchmarks, as well as the descriptive quality of their learned features in biochemical graph data exploration tasks. Our results show that LEGS-based networks match or outperforms popular GNNs, as well as the original geometric scattering construction, on many datasets, in particular in biochemical domains, while retaining certain mathematical properties of handcrafted (non-learned) geometric scattering.
KW - Geometric scattering
KW - graph neural networks
KW - graph signal processing
UR - http://www.scopus.com/inward/record.url?scp=85188503936&partnerID=8YFLogxK
U2 - 10.1109/TSP.2024.3378001
DO - 10.1109/TSP.2024.3378001
M3 - Article
AN - SCOPUS:85188503936
SN - 1053-587X
VL - 72
SP - 2939
EP - 2952
JO - IEEE Transactions on Signal Processing
JF - IEEE Transactions on Signal Processing
ER -