A General Graph Spectral Wavelet Convolution via Chebyshev Order Decomposition

Authors: Nian Liu, Xiaoxin He, Thomas Laurent, Francesco Di Giovanni, Michael M. Bronstein, Xavier Bresson

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments showcase the consistent improvements in both short-range and long-range tasks. This underscores the effectiveness of the proposed model in handling different scenarios. Our code is available at https: //github.com/liun-online/Wave GC.
Researcher Affiliation Collaboration 1National University of Singapore 2Loyola Marymount University 3University of Oxford 4AITHYRA, Austria. Correspondence to: Nian Liu <EMAIL>.
Pseudocode No The paper describes methods using mathematical formulations and textual explanations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https: //github.com/liun-online/Wave GC.
Open Datasets Yes Datasets for short-range tasks: CS, Photo, Computer and Cora Full from the Py Torch Geometric (Py G) (Fey & Lenssen, 2019), and one large-size graph, i.e. ogbn-arxiv from Open Graph Benchmark (OGB) (Hu et al., 2020) (2) Datasets for long-range tasks: Pascal VOC-SP (VOC), PCQM-Contact (PCQM), COCO-SP (COCO), Peptides-func (Pf) and Peptides-struct (Ps) from LRGB (Dwivedi et al., 2022).
Dataset Splits Yes For short-range (S) datasets, we follow the settings from (Chen et al., 2022). For ogbn-arxiv, we use the public splits in OGB (Hu et al., 2020). For longrange datasets, we adhere to the experimental configurations outlined in (Dwivedi et al., 2022).
Hardware Specification Yes GPU information: NVIDIA RTX A5000
Software Dependencies No We implement our Wave GC in Py Torch, and list some important parameter values in our model in Table 13. Please note that for the five long-range datasets, we follow the parameter budget 500k (Dwivedi et al., 2022). The paper mentions Py Torch but does not provide a specific version number for it or any other key software libraries.
Experiment Setup Yes Table 13. The values of parameters used in Wave GC (T: True; F: False). Dataset # parameters ρ J s Tight frames ℵ CS 495k 3 3 {0.5, 0.5, 0.5} T 0.1 Photo 136k 3 3 {1.0, 1.0, 1.0} T 0.1 Computer 167k 7 3 {10.0, 10.0, 10.0} T 0.1 Cora Full 621k 3 3 {2.0, 2.0, 2.0} T 0.1 ogbn-arxiv 2,354k 3 3 {5.0, 5.0, 5.0} F / Pascal VOC-SP 506k 5 3 {0.5, 1.0, 10.0} T / PCQM-Contact 508k 5 3 {0.5, 1.0, 5.0} T / COCO-SP 546k 3 3 {0.5, 1.0, 10.0} T / Peptides-func 496k 5 3 {10.0, 10.0, 10.0} T / Peptides-struct 534k 3 3 {10.0, 10.0, 10.0} F /