Research map / Graph learning / Graph self-supervised learning
Wiener Graph Deconvolutional Network Improves Graph Self-Supervised Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use WGDN rather than the strongest listed self-supervised baseline | Node classification · PubMed | accuracy: 81.0 $\pm$ 0.3 -> 81.9 $\pm$ 0.4 | Helped |
| Use WGDN rather than the strongest listed self-supervised baseline | Node classification · Computers, Photo, CS, and Physics | accuracy: 89.70 $\pm$ 0.15 -> 89.72 $\pm$ 0.48; 93.61 $\pm$ 0.20 -> 93.89 $\pm$ 0.31; 93.56 $\pm$ 0.15 -> 93.67 $\pm$ 0.14; 95.74 | Helped |
| Use WGDN rather than the strongest listed self-supervised baseline | Graph classification · IMDB-B, IMDB-M, PROTEINS, COLLAB, DD, and NCI1 | accuracy: 75.52 $\pm$ 0.66 -> 75.76 $\pm$ 0.20; 51.63 $\pm$ 0.52 -> 51.77 $\pm$ 0.55; 75.35 $\pm$ 0.09 -> 76.53 $\pm$ 0.38; 80.32 | Helped |
| Use WGDN rather than the strongest listed baseline | Node classification · Cora and OBGN-Arxiv | accuracy: 83.8 $\pm$ 0.5 -> 84.2 $\pm$ 0.6; 71.64 $\pm$ 0.12 -> 71.76 $\pm$ 0.23 | Helped |
| Use WGDN rather than CCA-SSG | Node classification · CiteSeer | accuracy: 73.1 $\pm$ 0.3 -> 72.2 $\pm$ 1.1 | Hurt |
| Add the graph Wiener decoder (ablation) | WGDN-AW · Five node- and graph-classification datasets (as discussed in ablation text) | accuracy: +6% relative improvement on PubMed | Helped |
| Add latent augmentation (ablation) | WGDN-A · Five node- and graph-classification datasets (as discussed in ablation text) | accuracy | Helped |
| Replace the Wiener decoder with an inverse decoder (ablation) | WGDN · Five node- and graph-classification datasets (as discussed in ablation text) | accuracy | Hurt |
| Use drop-edge or drop-node augmentation instead of latent Gaussian augmentation (ablation) | WGDN · PubMed, Computers, CS, IMDB-B, and DD | accuracy: WGDN-DE: 80.9 $\pm$ 0.6, 89.55 $\pm$ 0.36, 93.56 $\pm$ 0.31, 75.42 $\pm$ 0.15, 79.24 $\pm$ 0.40; WGDN-DN: 81.3 $\pm$ 0.5 | Mixed |
| Use a GCN filter rather than a heat filter (ablation) | WGDN · PubMed | accuracy: 81.9 -> 80.2 | Hurt |
| Use a PPR filter rather than a heat filter (ablation) | WGDN · PubMed | accuracy: 81.9 -> 81.4 | Hurt |
| Use heat rather than PPR or GCN propagation (ablation) | WGDN · Node classification datasets | accuracy | Helped |
Closest papers by meaning
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.86)
- Unifying Homophily and Heterophily for Spectral Graph Neural Networks via Triple Filter Ensembles (NeurIPS 2024 · similarity 0.86)
- Joint Graph Rewiring and Feature Denoising via Spectral Resonance (ICLR 2025 · similarity 0.86)
- Analysis of Corrected Graph Convolutions (NeurIPS 2024 · similarity 0.86)
- SLOG: An Inductive Spectral Graph Neural Network Beyond Polynomial Filter (ICML 2024 · similarity 0.85)
- Equivariant Machine Learning on Graphs with Nonlinear Spectral Filters (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.