Spectral Graph Pruning Against Over-Squashing and Over-Smoothing

Authors: Adarsh Jamadandi, Celia Rubio-Madrigal, Rebekka Burkholz

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments
Researcher Affiliation Academia 1Universität des Saarlandes 2CISPA Helmholtz Center for Information Security
Pseudocode Yes Algorithm 1 Proxy Spectral Gap based Greedy Graph Sparsification (PROXYDELETE)
Open Source Code Yes Our code https://github.com/Relational ML/Spectral Pruning Braess is available.
Open Datasets Yes We evaluate on the following datasets and tasks: 1) Pascal VOC-SP Semantic image segmentation as a node classification task operating on superpixel graphs. 2) Peptides-func Peptides modeled as molecular graphs. The task is graph classification. 3) Peptides-struct Peptides modeled as molecular graphs.
Dataset Splits Yes We use a 60/20/20 split for training/testing/validation respectively for all datasets.
Hardware Specification Yes All experiments were done on 2 V100 GPUs.
Software Dependencies No The paper mentions using "Py Torch Geometric and DGL library" but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We set the learning rate to {3e 3, 3e 4}, dropout to 0.32, and the hidden dimension size to 512. For GATs, the attention heads are set to 8.